Free
Meeting Abstracts  |   October 2005
Estimation of Errors in Determining Intrathoracic Blood Volume Using the Single Transpulmonary Thermal Dilution Technique in Hypovolemic Shock
Author Affiliations & Notes
  • Mahesh Nirmalan, M.D., F.R.C.A., Ph.D.
    *
  • Terrance M. Willard, M.Sc.
  • Dennis J. Edwards, F.R.C.P.
  • Rod A. Little, Ph.D.
    §
  • Paul M. Dark, F.R.C.S., Ph.D.
  • * Consultant, Critical Care Unit, Manchester Royal Infirmary. † Clinical Scientist, North Western Medical Physics Department, ‡ Director of Critical Care (retired), South Manchester University Hospitals, Manchester, United Kingdom. § Director, MRC Trauma Group, University of Manchester (retired), Manchester, United Kingdom. ∥ Consultant, Critical Care Unit Hope Hospital, Manchester, United Kingdom.
Article Information
Meeting Abstracts   |   October 2005
Estimation of Errors in Determining Intrathoracic Blood Volume Using the Single Transpulmonary Thermal Dilution Technique in Hypovolemic Shock
Anesthesiology 10 2005, Vol.103, 805-812. doi:
Anesthesiology 10 2005, Vol.103, 805-812. doi:
THERE is an increasing body of evidence showing that intrathoracic blood volume (ITBV) and global end-diastolic blood volume (GEDV) determined by the dye–thermal dilution technique provide a better estimate of cardiac preload than central venous pressure or pulmonary capillary wedge pressure.1–7 The dye–thermal dilution technique also provides an estimate of extravascular lung water (EVLW), which is a useful measure of early pulmonary edema.8,9 This technique involves the injection of cold indocyanine green dye (ICG) into the right atrium (or vena cava) followed by simultaneous recording of temperature (T) and dye–dilution curves in the abdominal aorta.10,11 The volume of distribution of temperature and ICG between the point of injection and sampling is a function of the mean transit time (MTt) of each indicator and cardiac output (CO). With improvements in technology, it is now possible to perform on-line measurement of temperature and ICG concentrations using a rapid response thermistor-tipped photometric catheter placed in the upper abdominal aorta. When this catheter is positioned above the origins of the major splanchnic and renal arteries, both indicators will distribute predominantly within the fluid compartments in the chest. While the thermal bolus distributes in the entire fluid compartment (intrathoracic thermal volume [ITTV]), ICG is confined to ITBV alone because of protein binding. Therefore ITTV, ITBV, and EVLW may be measured using the following steps12 :
However, the dye–thermal dilution technique, although effective, is expensive, time-consuming, and cumbersome for clinical use. This led to the development of the single transpulmonary thermodilution technique for the routine estimation of GEDV, ITBV, and EVLW in critically ill patients. The single transpulmonary thermal dilution technique uses the MTtTto derive ITTV as described in equation 1and the exponential down slope time of the thermodilution curve (DStT) to derive pulmonary thermal volume (PTV) and GEDV using the following steps13–15 :
Figure 1provides a schematic diagram summarizing the key stages of the dye–dilution technology.
Fig. 1. Schematic diagram of the relevant fluid compartments and their derivation. Note that it is feasible to derive global end-diastolic volume (GEDV) using the temperature dilution curves alone. CO = cardiac output; DSt = down slope time; ITBV = intrathoracic blood volume; ITTV = intrathoracic thermal volume; MTt = mean transit time; PTV = pulmonary thermal volume, which includes the pulmonary blood volume and the extravascular lung water. 
Fig. 1. Schematic diagram of the relevant fluid compartments and their derivation. Note that it is feasible to derive global end-diastolic volume (GEDV) using the temperature dilution curves alone. CO = cardiac output; DSt = down slope time; ITBV = intrathoracic blood volume; ITTV = intrathoracic thermal volume; MTt = mean transit time; PTV = pulmonary thermal volume, which includes the pulmonary blood volume and the extravascular lung water. 
Fig. 1. Schematic diagram of the relevant fluid compartments and their derivation. Note that it is feasible to derive global end-diastolic volume (GEDV) using the temperature dilution curves alone. CO = cardiac output; DSt = down slope time; ITBV = intrathoracic blood volume; ITTV = intrathoracic thermal volume; MTt = mean transit time; PTV = pulmonary thermal volume, which includes the pulmonary blood volume and the extravascular lung water. 
×
If ITBV is considered to be a constant proportion of GEDV, it is then possible to obtain an indirect estimate of ITBV and EVLW from the transpulmonary thermodilution curves alone. The commercially available device currently using this technology (PiCCO; PULSION; Medical Systems, Munich, Germany) uses a linear equation with a coefficient of 1.25 and an intercept of 0 to estimate ITBV from measured GEDV values:
This system has now been accepted for routine hemodynamic monitoring in many critical care units in the United Kingdom and in mainland Europe. The recent controversy related to the safety of pulmonary artery catheters in critically ill patients16 has no doubt encouraged this shift toward other forms of hemodynamic monitoring. (The acronym PiCCO points to the fact that this system provides a c  ontinuous estimate of c  ardiac o  utput using p  ulse contour analysis, and i  ntrathoracic blood volume and EVLW using the transpulmonary thermal dilution technique.) However, the derivation of ITBV using equation 6, a key step in the estimation of EVLW using the PiCCO system, has remained controversial ever since its introduction for measuring ITBV and EVLW in clinical monitoring. It has been argued that the relation between GEDV and ITBV may be influenced by overall volume status and CO. Furthermore, compensatory venous/arteriolar vasoconstriction in the pulmonary, systemic, and splanchnic circulations and the consequent redistribution of blood from the peripheral compartments to more central compartments may alter the numerical relation between GEDV and ITBV. These concerns have limited the acceptance of the PiCCO technology in many institutions. The current study was therefore undertaken to independently verify the numerical relation between GEDV and ITBV under normovolemic and severe hypovolemic conditions in a laboratory model of sustained shock and fluid resuscitation and to quantify the errors in using the PiCCO system under these conditions. We hypothesized that the regression equation describing the relation between GEDV and ITBV in normovolemic animals would lead to significant errors when used to estimate ITBV in animals with hypovolemic shock.
Materials and Methods
After institutional approval (Licence 42/1788; Home Office, Shrewsbury, United Kingdom), 19 immature female Large-White pigs (mean weight, 26.3 kg; SD, 3.3 kg) were randomly allocated to a control group (n = 9) and a shock group resuscitated with 4% succinylated gelatin (Maelor Pharmaceuticals Ltd., Wrexham, United Kingdom) (n = 10). Anesthesia was induced with halothane, oxygen, and nitrous oxide administered via  a snout mask followed by tracheal intubation and mechanical ventilation using a volume-cycled ventilator (Blease-Brompton-Manley; Chesham, Bucks, United Kingdom) (tidal volume, 10–15 ml/kg; rate, 12–15 breaths/min). An intravenous infusion of alphaxalone-alphadolone (Saffan; Pitman-Moore, Uxbridge, United Kingdom; 15 mg · kg−1· h−1) was commenced when venous access had been established. A pulmonary artery catheter (Baxter Swan-Ganz CCO/VIP, 7.5 French; Edwards Life Sciences, Irvine, CA) was sited via  the external jugular vein using aseptic techniques. All animals received maintenance fluids (0.9% NaCl, 10 ml ·b1 kg−1·b1 h−1) to replace insensible fluid loss. The dye–dilution catheter (Pulsiocath PV 2024, 4 French; PULSION) was positioned in the upper abdominal aorta via  the femoral route. Previous studies from our laboratory in pigs of similar proportions have shown that the distance from the femoral artery to the diaphragmatic crura was approximately 38 cm. The dye–dilution catheter was therefore advanced to 36–37 cm from the point of entry into the femoral artery to ensure that the tip of the catheter was positioned just below the level of the diaphragm. At the end of instrumentation, all animals were given a rest period of 30 min. After baseline measurements (time 0), the shock group was subjected to hemorrhagic shock by removing blood at a rate of 1 ml ·b1 kg−1·b1 min−1until hypovolemic shock was established. All hemodynamic measurements were repeated at this stage (30 min), and the animals were allowed to remain in shock for 60 min (90 min). During this shock phase, the presence of shock was confirmed using three of the following four predetermined endpoints:17 
  1. greater than 30% reduction in CO

  2. greater than 30% reduction in mean arterial pressure

  3. mixed venous oxygen saturation less than 40%

  4. blood lactate concentration greater than 3 mm

At the end of the shock phase, volume resuscitation was achieved using 4% succinylated gelatin. Fluid administration was stopped when CO had been restored and was maintained above 90% of baseline values. A final set of hemodynamic measurements were then made at 150 min. The animals were killed by anesthetic overdose, the lungs were removed, and EVLW content was determined by a gravimetric method that corrects for intravascular volume.18 
Measurement of GEDV, ITBV, and EVLWi
The dye–thermal dilution method (COLD Z-03; PULSION Medizintechnik, Munich, Germany) was used to measure GEDV, measured ITBV (ITBVm), and extravascular lung water index (EVLWi) at 0, 30, 90, and 150 min. Duplicate estimates of EVLWi using a manual injection of 10 ml cold ICG (PULSION; 1 mg/ml) were made, and the numerical average of the two closest measurements was taken as the true EVLWi. If the difference in EVLWi between the duplicate injections was greater than 10%, a third injection was made in keeping with current clinical practice, and the average of two closest values was used in all subsequent calculations. At each of the four time points, other hemodynamic variables, including heart rate, pulmonary artery thermodilution, transpulmonary thermodilution, and mean arterial pressure were also recorded.
Cardiac Output and Mean Arterial Pressure
The Vigilance continuous CO monitor (Baxter Healthcare Ltd., Deerfield, IL) was used to measure continuous CO and intermittent thermodilution CO. Continuous CO measurements were stopped at the four time points when cold ICG was injected for CO, GEDV, and ITBVmmeasurements. Continuous CO measurements were used to control the volume of blood loss and adequacy of fluid resuscitation only and were not used in any of the subsequent statistical analyses. Arterial pressure was transduced directly from the side arm of the aortic cannula. The pressure signals were acquired and stored in a personal computer using standard signal processing equipment and software (CED 1902, CED1401 and Spike 2; Cambridge Electronics Design, Cambridge, United Kingdom).
Statistical Analyses
The relation between GEDV and ITBVmin control and shock groups was first estimated using linear regression analyses where the correlation coefficients and y-intercepts were compared using 95% confidence intervals (CIs). Hemodynamic variables were analyzed using analysis of variance for repeated measurements (general linear model; SPSS 9.0; SPSS Inc., Chicago, IL). Significant factors were further compared using 95% CIs of estimated means at each of the four stages. Statistical significance was defined as P  < 0.05 (two sided). Bland-Altman plots and within-subject correlation were used to compare the different measures of CO and measured/estimated ITBV.19,20 Because all data were normally distributed, mean (SD) values were used as summary statistics.
Derivation of Estimated ITBV and Percent Estimation Error
Because CO is a common factor in the derivation of ITBVmand GEDV, the regression plots are likely to be influenced by mathematical coupling between these two parameters. An alternative approach was therefore also used in data analysis whereby estimated ITBV (ITBVe: ITBV estimated indirectly using a regression equation) was compared against ITBVm, and prediction errors at the four stages were compared using repeated-measures analysis of variance. The linear regression equation estimated using 32 pairs of measured ITBVmand GEDV values from 8 animals in the control group was applied to the 4 GEDV measurements in the ninth animal to obtain the corresponding “estimated ITBV” (ITBVe) for the ninth animal in the control group. This process was repeated for each of the nine animals in the group, allowing a total of 36 comparisons between ITBVmand ITBVewithin the control group. The regression equation developed using this technique is not influenced by data from the animal in which the equation would be put to use. This out-of-sample prediction technique is in the spirit of the “leave-one-out” cross-validation technique.121 The equation from all the GEDV/ITBVmmeasurements from the entire control group (36 pairs of measurements; 9 animals and 4 sets of readings per animal) was then applied to the GEDV measurements in the shock group to obtain the corresponding ITBVefor each of the animal in the shock group. The difference between ITBVmand ITBVewas expressed as a percentage of ITBVm([(ITBVm− ITBVe)/ITBVm]× 100) and used as the percent prediction error at each of the time points. When this approach is adopted, the quantitative relation between ITBVmand GEDV under normovolemic conditions is imposed on the shock group even at 30 and 90 min when the animals were in shock. The percent errors for the shock group at 30 and 90 min were therefore corrected using a second set of regression equations estimated from ITBVm/GEDV values from the shock group at 30 and 90 min only. Values from 9 animals in the shock group (18 pairs of GEDV/ITBVmat 30 and 90 min) were used to derive the regression equation to be used in the tenth animal (and the process repeated for all 10 animals) in keeping with the out-of-sample prediction technique.21 Finally, the PiCCO algorithm (equation 6) was used to obtain ITBVPiCCOin both groups of animals and compared against ITBVmusing Bland-Altman plots. Percent errorPiCCOwas defined as [(ITBVm− ITBVPiCCO)/ITBVm]× 100.
Results
The mean weights for both groups were similar (control group, 26.8 [3.0] kg; shock group, 27.8 [3.8] kg). Gravimetrically determined EVLWi data were available in only 15 animals and ranged between 5.9 and 12.2 ml/kg (mean, 8.4 ml/kg; SD, 2.1 ml/kg). In these animals, the percentage EVLW detected by the dye–thermal dilution technique was approximately 80% of lung water determined by the gravimetric method. In the shock group, the shock phase was associated with a significant reduction in CO, stroke volume, and mean arterial pressure at 30 and 90 min (P  < 0.05). The above changes were, as expected, accompanied by a significant reduction (P  < 0.05) in GEDV indexed to body weight (GEDVi) and ITBVmindexed to body weight (ITBVim). The hemodynamic variables for both groups during the entire experiment are summarized in table 1. The relation between CO measurements obtained by pulmonary artery thermodilution and transpulmonary thermodilution in shock is of considerable clinical interest and is summarized in figure 2.
Table 1. Hemodynamic Variables for Control and Shock Groups 
Image not available
Table 1. Hemodynamic Variables for Control and Shock Groups 
×
Fig. 2. Comparison between cardiac output (CO; l/min) measurements obtained by pulmonary artery thermodilution (COpa) and transpulmonary thermodilution (COaort) using within-subject correlation (  A  ) and Bland-Altman plot (  B  ) (within-subject correlation  r  2= 0.95; bias =−0.21 and SD = 0.51). The bias of −0.21 indicates that transpulmonary thermodilution systematically overestimates CO.  Open triangles  = time 0;  closed triangles  = 30 min;  closed squares  = 90 min;  open squares  = 150 min. 
Fig. 2. Comparison between cardiac output (CO; l/min) measurements obtained by pulmonary artery thermodilution (COpa) and transpulmonary thermodilution (COaort) using within-subject correlation (  A  ) and Bland-Altman plot (  B  ) (within-subject correlation  r  2= 0.95; bias =−0.21 and SD = 0.51). The bias of −0.21 indicates that transpulmonary thermodilution systematically overestimates CO.  Open triangles  = time 0;  closed triangles  = 30 min;  closed squares  = 90 min;  open squares  = 150 min. 
Fig. 2. Comparison between cardiac output (CO; l/min) measurements obtained by pulmonary artery thermodilution (COpa) and transpulmonary thermodilution (COaort) using within-subject correlation (  A  ) and Bland-Altman plot (  B  ) (within-subject correlation  r  2= 0.95; bias =−0.21 and SD = 0.51). The bias of −0.21 indicates that transpulmonary thermodilution systematically overestimates CO.  Open triangles  = time 0;  closed triangles  = 30 min;  closed squares  = 90 min;  open squares  = 150 min. 
×
The correlation between GEDV and ITBVmfor control and shock groups are summarized in figure 3. The regression equation obtained using all the 36 pairs of measurements in the control group was
Fig. 3. (  A  ) Correlation plot showing the relation between end-diastolic blood volume (GEDV) and measured intrathoracic blood volume (ITBVm) in the control group (ITBVm= 1.21 × GEDV + 99;  r  2= 0.89,  P  < 0.0001; 95% confidence interval [CI] for slope, 1.1–1.4 and 95% CI for y-intercept, 31–168). (  B  ) Shock group under normovolemic (  open triangles and squares  with the correlation line shown as  interrupted line  : ITBVm= 1.3 × GEDV + 58;  r  2= 0.92,  P  < 0.0001; 95% CI for slope, 1.1 to 1.5 and 95% CI for y-intercept, −33 to 150) and hypovolemic conditions (  shaded triangles and squares  with the correlation line shown as a  solid line  : ITBVm= 1.45 × GEDV + 0.6;  r  2= 0.95,  P  < 0.0001; 95% CI for slope, 1.3 to 1.6 and 95% CI for y-intercept, −47 to 49). The differences between the slopes and intercepts of the three equations are not statistically significant. 
Fig. 3. (  A  ) Correlation plot showing the relation between end-diastolic blood volume (GEDV) and measured intrathoracic blood volume (ITBVm) in the control group (ITBVm= 1.21 × GEDV + 99;  r  2= 0.89,  P  < 0.0001; 95% confidence interval [CI] for slope, 1.1–1.4 and 95% CI for y-intercept, 31–168). (  B  ) Shock group under normovolemic (  open triangles and squares  with the correlation line shown as  interrupted line  : ITBVm= 1.3 × GEDV + 58;  r  2= 0.92,  P  < 0.0001; 95% CI for slope, 1.1 to 1.5 and 95% CI for y-intercept, −33 to 150) and hypovolemic conditions (  shaded triangles and squares  with the correlation line shown as a  solid line  : ITBVm= 1.45 × GEDV + 0.6;  r  2= 0.95,  P  < 0.0001; 95% CI for slope, 1.3 to 1.6 and 95% CI for y-intercept, −47 to 49). The differences between the slopes and intercepts of the three equations are not statistically significant. 
Fig. 3. (  A  ) Correlation plot showing the relation between end-diastolic blood volume (GEDV) and measured intrathoracic blood volume (ITBVm) in the control group (ITBVm= 1.21 × GEDV + 99;  r  2= 0.89,  P  < 0.0001; 95% confidence interval [CI] for slope, 1.1–1.4 and 95% CI for y-intercept, 31–168). (  B  ) Shock group under normovolemic (  open triangles and squares  with the correlation line shown as  interrupted line  : ITBVm= 1.3 × GEDV + 58;  r  2= 0.92,  P  < 0.0001; 95% CI for slope, 1.1 to 1.5 and 95% CI for y-intercept, −33 to 150) and hypovolemic conditions (  shaded triangles and squares  with the correlation line shown as a  solid line  : ITBVm= 1.45 × GEDV + 0.6;  r  2= 0.95,  P  < 0.0001; 95% CI for slope, 1.3 to 1.6 and 95% CI for y-intercept, −47 to 49). The differences between the slopes and intercepts of the three equations are not statistically significant. 
×
(r  2= 0.89, P  < 0.0001; 95% CI for slope, 1.1–1.4 and 95% CI for y-intercept, 31–168). The regression equation obtained in the shock group under normovolemic conditions (at 0 and 150) minutes was
(r  2= 0.92, P  < 0.0001; 95% CI for slope, 1.1 to 1.5 and 95% CI for y-intercept, −33 to 150). The regression equation obtained in the shock group under hypovolemic conditions at 30 and 90 min was
(r  2= 0.95, P  < 0.0001; 95% CI for slope, 1.3 to 1.6 and 95% CI for y-intercept, −47 to 49). The 95% CIs for the y-intercept were very wide, and consequently the differences between the y-intercepts in equations 7, 8, and 9were not statistically significant (P  > 0.05). The family of regression equations developed within the control group using the leave-one-out cross-validation strategy provided an accurate estimate of ITBV in the control group (mean bias, 0.9%; SD, 4.2%), and the mean percent prediction error was less than 5% at all four time points (table 1). In the shock group, however, equation 7resulted in overestimation of ITBV at 30 and 90 min, and consequently, the percent prediction error was significantly greater (P  < 0.05) than the corresponding errors at 0 and 150 min (table 1). However, the family of equations developed using the shock-phase GEDV/ITBVmvalues only provided a more accurate estimate of ITBV in the shock group at 30 and 90 min (mean bias, 5.4%; SD, 4.6%). The distribution of errors at the four time points for both groups are summarized in figure 4. Even though the PiCCO equation significantly underestimated ITBV (P  < 0.01; control group: mean bias, 12.9%; SD, 4.3%; shock group: mean bias, 12.2%; SD, 3.9%), the percent prediction errors were similar and less than 15% in both groups at all four stages of the experiment (table 1).
Fig. 4. (  A  ) Distribution of percent prediction error in estimating intrathoracic blood volume (ITBV) in group C using  equation 7. (  B  ) Distribution of percent error in estimating ITBVefor group S showing the significant increase in error at 30 and 90 min (  open triangles  = time 0;  shaded triangles  = 30 min;  shaded squares  = 90 min;  open squares  = 150 min). (  C  ) Percent error in group S at 30 and 90 min using  equation 7compared with “corrected errors” when ITBVewas estimated using  equation 9(  shaded triangles  = errors at 30 min;  dotted triangles  = corrected errors at 30 min;  shaded squares  = errors at 90 min;  dotted squares  = corrected errors at 90 min). Percent prediction error refers to [(ITBVm− ITBVe)/ITBVm]× 100. 
Fig. 4. (  A  ) Distribution of percent prediction error in estimating intrathoracic blood volume (ITBV) in group C using  equation 7. (  B  ) Distribution of percent error in estimating ITBVefor group S showing the significant increase in error at 30 and 90 min (  open triangles  = time 0;  shaded triangles  = 30 min;  shaded squares  = 90 min;  open squares  = 150 min). (  C  ) Percent error in group S at 30 and 90 min using  equation 7compared with “corrected errors” when ITBVewas estimated using  equation 9(  shaded triangles  = errors at 30 min;  dotted triangles  = corrected errors at 30 min;  shaded squares  = errors at 90 min;  dotted squares  = corrected errors at 90 min). Percent prediction error refers to [(ITBVm− ITBVe)/ITBVm]× 100. 
Fig. 4. (  A  ) Distribution of percent prediction error in estimating intrathoracic blood volume (ITBV) in group C using  equation 7. (  B  ) Distribution of percent error in estimating ITBVefor group S showing the significant increase in error at 30 and 90 min (  open triangles  = time 0;  shaded triangles  = 30 min;  shaded squares  = 90 min;  open squares  = 150 min). (  C  ) Percent error in group S at 30 and 90 min using  equation 7compared with “corrected errors” when ITBVewas estimated using  equation 9(  shaded triangles  = errors at 30 min;  dotted triangles  = corrected errors at 30 min;  shaded squares  = errors at 90 min;  dotted squares  = corrected errors at 90 min). Percent prediction error refers to [(ITBVm− ITBVe)/ITBVm]× 100. 
×
Discussion
In the current study, we used the volume of distribution of ICG between the right atrium and upper abdominal aorta (immediately below the level of the diaphragm) as the standard measure of ITBV.10,11 We compared alternate measures of ITBV, i.e.  , ITBVederived indirectly as a linear function of calculated GEDV, under normovolemic and hypovolemic conditions against this standard measure to estimate prediction errors. Our results confirm that the linear relation between GEDV and ITBV is preserved in hypovolemic shock (fig. 3; r  2= 0.95, P  < 0.0001). However, the precise numerical relation and the corresponding regression equations relating to the two variables are influenced by circulatory volume and CO as illustrated by the wide 95% CI in the y-intercepts for equations 7, 8, and 9(fig. 3). The significant increase in prediction errors when ITBVeduring the shock phase was derived from the equation developed in the control group under normovolemic conditions (table 1) confirms this inherent dependence on circulatory volume and CO. The clinical relevance of this small but statistically significant effect of circulatory volume/CO on the numerical relation between GEDV and ITBVmand its relevance to clinical monitoring using the PiCCO system requires further clarification.
Although the prediction errors using the PiCCO equation were significantly greater than with equation 7 or 9, the overall errors were less than 15% in both groups (table 1). The correlation coefficient of 1.25 in the PiCCO algorithm is between the two coefficients determined in the two study groups (equations 7 or 9, control group: 1.21; shock group: 1.45). Furthermore, by not incorporating the y-intercept, the PiCCO equation (ITBV = 1.25 × GEDV) eliminates one of the main sources of variation related to circulatory volume/CO. Consequently, the mean bias was similar in the control and shock groups (control: mean bias, 12.9%; SD, 4.3%; shock: normovolemia; mean bias, 10.7%; SD, 3.2%; hypovolemia; mean bias, 12.2%; SD, 3.9%), and no significant changes in percent prediction error were seen in the shock group during the four stages of the experiment (table 1). Because uncertainties due to extraneous factors such as CO or circulatory volume are usually more important in clinical monitoring, the strategy adopted by the PiCCO system seems sound and clinically meaningful.
Three potential limitations of the current study require emphasis. First, because CO is a common parameter in the derivation of GEDV and ITBV, the two measurements are mathematically coupled. The effects of mathematical coupling are bound to distort any observation based on a direct comparison between the regression equations when the common factor (CO) is subject to major changes during the course of the study. Our conclusions are therefore based primarily on “prediction errors” on comparing measured and estimated ITBV. In this context, the leave-one-out cross-validation technique provides a robust strategy by ensuring that the regression equation applied to any given animal is not influenced by values from the same animal.1Second, the relation between GEDV and ITBV was evaluated under normovolemic and extreme hypovolemic conditions. We estimated blood volume on the basis of 75 ml/kg based on previous studies in our laboratory22 and the volume of blood removed from each animal (mean, 39%; SD, 11% of estimated blood volume) varied to achieve the predetermined endpoints of shock. This degree of hypovolemia is likely to be recognized and corrected before EVLW measurements become relevant in clinical management. The fact that, even under such extreme conditions, the PiCCO algorithm was robust enough to predict ITBV with an overall error of less than 15% is reassuring. Third, volume of distribution measured using the dye–dilution technique refers to the respective volume from the point of injection (right atrium or superior vena cava) to the point of sampling (upper abdominal aorta just below the level of the diaphragm).10,11,13 Consequently the term global end-diastolic volume  is a misnomer because it does not equate to the volume of blood in the four cardiac chambers alone. Nevertheless, this term has been used extensively in published literature, and we have retained its use in our study to maintain consistency. The basic premise, however, is that ITBV can be derived indirectly from a closely related central blood volume. Whether this central blood volume should be referred to as global end-diastolic volume, as suggested in current literature and by the manufacturer (PiCCO; PULSION; Medical Systems, Munich, Germany), remains controversial and should be addressed by an appropriate consensus group.
The study also shows that CO determined by transpulmonary thermodilution compared favorably with pulmonary artery thermodilution technique during the four stages of the experiment (fig. 2). These findings therefore confirm the view that transpulmonary thermodilution technique provides a reliable alternative to the pulmonary artery thermodilution technique to measure CO. Previous investigators have reported similar findings in other clinical conditions such as subarachnoid hemorrhage, sepsis, and major burns and in patients undergoing heart transplantation.3,23,24 Our findings in a large animal model of hemorrhagic shock add to the current body of evidence in this area. Estimates of EVLW based on the single thermal dilution technique have been shown to be significantly different from the corresponding estimates based on dye–thermal dilution technique in a previous clinical study that involved a small group of patients with septic shock.25 However, this early study used an older technology based on off-line measurement of ICG using an external dye densitometer. Wickerts et al.  26 have shown that off-line measurement of ICG concentration can introduce significant errors in the estimation of EVLW due to phase delays between temperature and ICG–dilution curves. In a more recent study in patients with major burns (mean body surface area, 46%; range, 26–67%), Kuntscher et al.  27 have shown that ITBV estimated by the single thermal dilution technique has only a weak correlation (r  = 0.37) with ITBV determined by the dye–thermal dilution technique. It is widely acknowledged that uniform pulmonary perfusion is an essential prerequisite for the estimation of EVLW using the dye–dilution principles. Using an experimental model of regional pulmonary hypoperfusion, Schreiber et al.  28 have shown that estimates of GEDV and ITBV may also be perfusion dependant and suggested that an increase in mean pulmonary transit velocity due to vasoconstriction may lead to a reduction in transit time and consequently underestimation of GEDV and ITBV. None of the above three studies,25,27,28 however, provide any quantitative data on the numerical relation between GEDV and ITBV in hypovolemic states. In a previous study, we have shown that the numerical relation between GEDV and ITBV was not affected significantly by the presence of acute lung injury and small changes (< 10%) in total blood volume.15 We have explored this issue further in the current study and have demonstrated that although the presence of larger volume deficits does influence this relation between GEDV and ITBV significantly, the resultant errors were generally small (< 15%) and within clinically tolerable limits. These findings should be taken into account when the single thermal dilution technique is applied for research purposes where a more accurate estimate of ITBV may be required. The limited nature of the current study and the relatively small sample size unfortunately precludes any formal subgroup analyses to identify cohorts of animals where prediction errors exceed 15%. We believe that this issue can be addressed meaningfully only through a larger clinical study dealing with a mixed critically ill population. It is also necessary to point out that the applications of the PiCCO system extend far beyond the estimation of ITBV/EVLW alone, and these other facets of the PiCCO technology are beyond the scope of our work and have not been commented on.
In summary, this study demonstrates that the linear relation between GEDV and ITBV is maintained in severe hypovolemic shock. Even though the exact numerical relation between GEDV and ITBV is influenced by CO and circulatory volume, the overall errors in predicting ITBV from measured GEDV were small and within clinically tolerable limits. The correlation coefficient of 1.25 and an intercept of 0 used in the PiCCO algorithm overcomes some of the variations related to CO and circulatory volume and consequently provides a relatively robust clinical measure of ITBV and EVLW.
The authors thank Tim Riney, B.Sc., and Hazel Marshall, B.Sc. (Senior Technicians, MRC Trauma Group, University of Manchester, Manchester, United Kingdom), for their help in conducting these experiments, and Malachy O. Columb, F.R.C.A. (Consultant Anesthesiologist, South Manchester University Hospitals, Withenshaw, Manchester, United Kingdom), for statistical advice.
References
Kumar A, Anel R, Bunnell E, Habet K, Zanotti S, Marshall S, Neumann A, Ali A, Cheang M, Kavinsky C, Parrillo JE: Pulmonary artery occlusion pressure and central venous pressure fail to predict ventricular filling volume, cardiac performance or the response to volume infusion in normal subjects. Crit Care Med 2004; 32:691–9Kumar, A Anel, R Bunnell, E Habet, K Zanotti, S Marshall, S Neumann, A Ali, A Cheang, M Kavinsky, C Parrillo, JE
Rocca GD, Costa GM, Coccia C, Pompei L, Di Marco P, Pietropaoli P: Preload index: Pulmonary artery occlusion pressure versus intrathoracic blood volume monitoring in lung transplantation. Anesth Analg 2002; 95:835–43Rocca, GD Costa, GM Coccia, C Pompei, L Di Marco, P Pietropaoli, P
Holm C, Melcer B, Horbrand F, Worl HH, von Donnersmarck GH, Muhlbauer W: Intrathoracic blood volume as an end-point in resuscitation of the severely burned: an observational study of 24 patients. J Trauma 2000; 48:728–34Holm, C Melcer, B Horbrand, F Worl, HH von Donnersmarck, GH Muhlbauer, W
Goedje O, Seebauer T, Peyerl M, Pfeiffer UJ, Reichart B: Hemodynamic monitoring by double-indicator dilution technique in patients after orthotopic heart transplantation. Chest 2000; 118:775–81Goedje, O Seebauer, T Peyerl, M Pfeiffer, UJ Reichart, B
Goedje O, Peyerl M, Seebauer T, Lamm P, Mair H, Reichart B: Central venous pressure, pulmonary capillary wedge pressure and intrathoracic blood volumes as preload indicators in cardiac surgery patients. E J Cardiothoracic surgery 1998; 13:533–40Goedje, O Peyerl, M Seebauer, T Lamm, P Mair, H Reichart, B
Hoeft A, Schorn B, Weyland A, Scholz M, Buhre W, Stepanek E, Sonntag H: Bedside assessment of intravascular volume status in patients undergoing coronary bypass surgery. Anesthesiology 1994; 81:76–86Hoeft, A Schorn, B Weyland, A Scholz, M Buhre, W Stepanek, E Sonntag, H
Lichtwarck-Aschoff M, Zeravik J, Pfeiffer UJ: Intrathoracic blood volume accurately reflects circulatory volume status in critically ill patients with mechanical ventilation. Intensive Care Med 1992; 18:142–7Lichtwarck-Aschoff, M Zeravik, J Pfeiffer, UJ
Halperin BD, Feeley TW, Mihm FG, Chiles C, Guthaner DF, Blank NE: Evaluation of the portable chest roentgenogram for quantifying extravascular lung water in critically ill adults. Chest 1985; 88:649–52Halperin, BD Feeley, TW Mihm, FG Chiles, C Guthaner, DF Blank, NE
Baudendistel L, Shields JB, Kaminski DL: Comparison of double indicator thermodilution measurements of extravascular lung water (EVLW) with radiographic estimation of lung water in trauma patients. J Trauma 1982; 22:983–8Baudendistel, L Shields, JB Kaminski, DL
Lewis FR, Elings VB, Hill SL, Christensen JM: The measurement of extravascular lung water by thermal-green dye indicator dilution. Ann N Y Acad Sci 1982; 384:394–410Lewis, FR Elings, VB Hill, SL Christensen, JM
Mihm FG, Feeley TW, Rosenthal MH, Lewis F: Measurement of extravascular lung water in dogs using the thermal-green dye indicator dilution method. Anesthesiology 1982; 57:116–22Mihm, FG Feeley, TW Rosenthal, MH Lewis, F
Goedje O, Peyerl M, Seebauer T, Dewald O, Reichart B: Reproducibility of double indicator dilution measurements of intrathoracic blood volume compartments, extravascular lung water and liver functions. Chest 1998; 113:1070–7Goedje, O Peyerl, M Seebauer, T Dewald, O Reichart, B
Sakka SG, Ruhl CC, Pfeiffer UJ, Beale R, McLuckie A, Reinhart K, Meier-Hellmann A: Assessment of cardiac preload and extravascular lung water by single transpulmonary thermodilution. Intensive Care Med 2000; 26:180–7Sakka, SG Ruhl, CC Pfeiffer, UJ Beale, R McLuckie, A Reinhart, K Meier-Hellmann, A
Newman P: Extravascular lung water and intrathoracic blood volume: Double versus single indicator dilution technique. Intensive Care Med 1999; 25:216–9Newman, P
Nirmalan M, Niranjan M, Willard T, Edwards JD, Little RA, Dark PM: Estimation of errors in determining intrathoracic blood volume using thermal dilution in pigs with acute lung injury and haemorrhage. Br J Anaesth 2004; 93:546–51Nirmalan, M Niranjan, M Willard, T Edwards, JD Little, RA Dark, PM
Connors AF, Speroff T, Dawson NV, Thomas C, Harrell FE, Wagner D, Desbiens N, Goldman L, Wu AW, Califf RM, Fulkerson WJ, Vidaillet H, Broste S, Bellamy P, Lynn J, Knaus WA: The effectiveness of right heart catheterisation in the initial care of critically ill patients. SUPPORT Investigators. J Am Med Assoc 1996; 276:889–97Connors, AF Speroff, T Dawson, NV Thomas, C Harrell, FE Wagner, D Desbiens, N Goldman, L Wu, AW Califf, RM Fulkerson, WJ Vidaillet, H Broste, S Bellamy, P Lynn, J Knaus, WA
Nirmalan M, Willard T, Edwards JD, Dark P, Little RA: Effects of sustained post-traumatic shock and initial fluid resuscitation on extravascular lung water content and pulmonary vascular pressures in a porcine model of shock. Br J Anaesth 2003; 91:224–32Nirmalan, M Willard, T Edwards, JD Dark, P Little, RA
Pearce ML, Yamashita J, Beazell J: Measurement of pulmonary oedema. Circ Res 1965; 16:482–8Pearce, ML Yamashita, J Beazell, J
Bland JM, Altman DG: Calculating correlation coefficients with repeated observations: I. Correlation within subjects. BMJ 1995; 310:446Bland, JM Altman, DG
Bland JM, Altman DG: Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1986; 1:307–10Bland, JM Altman, DG
Bishop CM: Learning and generalisation, Neural Networks for Pattern Recognition. New York, Oxford University Press, 1995, pp 372–5Bishop, CM Learning and generalisation, New York Oxford University Press
Foex BA, Kirkman E, Little RA: Injury (nociceptive afferent nerve stimulation) modifies the hemodynamic and metabolic responses to hemorrhage in immature swine. Crit Care Med 2004; 32:740–6Foex, BA Kirkman, E Little, RA
Sakka SG, Reinhart K, Meier-Hellmann A: Comparison of pulmonary artery catheter and arterial thermodilution cardiac output in critically ill patients. Intensive Care Med 1999; 25:843–6Sakka, SG Reinhart, K Meier-Hellmann, A
Holm C, Horbrand MF, von Donnersmarck H, Muhlbauer W: Arterial thermodilution: An alternative to pulmonary artery catheter for cardiac output assessment in burn patients. Burns 2001; 27:161–6Holm, C Horbrand, MF von Donnersmarck, H Muhlbauer, W
Schuster DP, Calandrino FS: Single versus double indicator dilution measurements of extravascular lung water. Crit Care Med 1991; 19:84–8Schuster, DP Calandrino, FS
Wickerts CJ, Jakobsson J, Frostell C, Hedenstierna G: Measurement of extravascular lung water by thermal-dye dilution technique: Mechanisms of cardiac output dependence. Intensive Care Med 1990; 16:115–20Wickerts, CJ Jakobsson, J Frostell, C Hedenstierna, G
Kuntscher MV, Czermak C, Blome-Eberwein S, Dacho A, Germann G: Transpulmonary thermal dye versus single thermodilution methods for assessment of intrathoracic blood volume and extravascular lung water in major burns resuscitation. J Burns Care Rehabil 2003; 24:142–7Kuntscher, MV Czermak, C Blome-Eberwein, S Dacho, A Germann, G
Schreiber T, Huter L, Schwarzkopf K, Schubert H, Preussler N, Bloos F, Gaser E, Karzai W: Lung perfusion affects preload assessment and lung water calculation with the transpulmonary double indicator method. Intensive Care Med 2001; 27:1814–8Schreiber, T Huter, L Schwarzkopf, K Schubert, H Preussler, N Bloos, F Gaser, E Karzai, W
Fig. 1. Schematic diagram of the relevant fluid compartments and their derivation. Note that it is feasible to derive global end-diastolic volume (GEDV) using the temperature dilution curves alone. CO = cardiac output; DSt = down slope time; ITBV = intrathoracic blood volume; ITTV = intrathoracic thermal volume; MTt = mean transit time; PTV = pulmonary thermal volume, which includes the pulmonary blood volume and the extravascular lung water. 
Fig. 1. Schematic diagram of the relevant fluid compartments and their derivation. Note that it is feasible to derive global end-diastolic volume (GEDV) using the temperature dilution curves alone. CO = cardiac output; DSt = down slope time; ITBV = intrathoracic blood volume; ITTV = intrathoracic thermal volume; MTt = mean transit time; PTV = pulmonary thermal volume, which includes the pulmonary blood volume and the extravascular lung water. 
Fig. 1. Schematic diagram of the relevant fluid compartments and their derivation. Note that it is feasible to derive global end-diastolic volume (GEDV) using the temperature dilution curves alone. CO = cardiac output; DSt = down slope time; ITBV = intrathoracic blood volume; ITTV = intrathoracic thermal volume; MTt = mean transit time; PTV = pulmonary thermal volume, which includes the pulmonary blood volume and the extravascular lung water. 
×
Fig. 2. Comparison between cardiac output (CO; l/min) measurements obtained by pulmonary artery thermodilution (COpa) and transpulmonary thermodilution (COaort) using within-subject correlation (  A  ) and Bland-Altman plot (  B  ) (within-subject correlation  r  2= 0.95; bias =−0.21 and SD = 0.51). The bias of −0.21 indicates that transpulmonary thermodilution systematically overestimates CO.  Open triangles  = time 0;  closed triangles  = 30 min;  closed squares  = 90 min;  open squares  = 150 min. 
Fig. 2. Comparison between cardiac output (CO; l/min) measurements obtained by pulmonary artery thermodilution (COpa) and transpulmonary thermodilution (COaort) using within-subject correlation (  A  ) and Bland-Altman plot (  B  ) (within-subject correlation  r  2= 0.95; bias =−0.21 and SD = 0.51). The bias of −0.21 indicates that transpulmonary thermodilution systematically overestimates CO.  Open triangles  = time 0;  closed triangles  = 30 min;  closed squares  = 90 min;  open squares  = 150 min. 
Fig. 2. Comparison between cardiac output (CO; l/min) measurements obtained by pulmonary artery thermodilution (COpa) and transpulmonary thermodilution (COaort) using within-subject correlation (  A  ) and Bland-Altman plot (  B  ) (within-subject correlation  r  2= 0.95; bias =−0.21 and SD = 0.51). The bias of −0.21 indicates that transpulmonary thermodilution systematically overestimates CO.  Open triangles  = time 0;  closed triangles  = 30 min;  closed squares  = 90 min;  open squares  = 150 min. 
×
Fig. 3. (  A  ) Correlation plot showing the relation between end-diastolic blood volume (GEDV) and measured intrathoracic blood volume (ITBVm) in the control group (ITBVm= 1.21 × GEDV + 99;  r  2= 0.89,  P  < 0.0001; 95% confidence interval [CI] for slope, 1.1–1.4 and 95% CI for y-intercept, 31–168). (  B  ) Shock group under normovolemic (  open triangles and squares  with the correlation line shown as  interrupted line  : ITBVm= 1.3 × GEDV + 58;  r  2= 0.92,  P  < 0.0001; 95% CI for slope, 1.1 to 1.5 and 95% CI for y-intercept, −33 to 150) and hypovolemic conditions (  shaded triangles and squares  with the correlation line shown as a  solid line  : ITBVm= 1.45 × GEDV + 0.6;  r  2= 0.95,  P  < 0.0001; 95% CI for slope, 1.3 to 1.6 and 95% CI for y-intercept, −47 to 49). The differences between the slopes and intercepts of the three equations are not statistically significant. 
Fig. 3. (  A  ) Correlation plot showing the relation between end-diastolic blood volume (GEDV) and measured intrathoracic blood volume (ITBVm) in the control group (ITBVm= 1.21 × GEDV + 99;  r  2= 0.89,  P  < 0.0001; 95% confidence interval [CI] for slope, 1.1–1.4 and 95% CI for y-intercept, 31–168). (  B  ) Shock group under normovolemic (  open triangles and squares  with the correlation line shown as  interrupted line  : ITBVm= 1.3 × GEDV + 58;  r  2= 0.92,  P  < 0.0001; 95% CI for slope, 1.1 to 1.5 and 95% CI for y-intercept, −33 to 150) and hypovolemic conditions (  shaded triangles and squares  with the correlation line shown as a  solid line  : ITBVm= 1.45 × GEDV + 0.6;  r  2= 0.95,  P  < 0.0001; 95% CI for slope, 1.3 to 1.6 and 95% CI for y-intercept, −47 to 49). The differences between the slopes and intercepts of the three equations are not statistically significant. 
Fig. 3. (  A  ) Correlation plot showing the relation between end-diastolic blood volume (GEDV) and measured intrathoracic blood volume (ITBVm) in the control group (ITBVm= 1.21 × GEDV + 99;  r  2= 0.89,  P  < 0.0001; 95% confidence interval [CI] for slope, 1.1–1.4 and 95% CI for y-intercept, 31–168). (  B  ) Shock group under normovolemic (  open triangles and squares  with the correlation line shown as  interrupted line  : ITBVm= 1.3 × GEDV + 58;  r  2= 0.92,  P  < 0.0001; 95% CI for slope, 1.1 to 1.5 and 95% CI for y-intercept, −33 to 150) and hypovolemic conditions (  shaded triangles and squares  with the correlation line shown as a  solid line  : ITBVm= 1.45 × GEDV + 0.6;  r  2= 0.95,  P  < 0.0001; 95% CI for slope, 1.3 to 1.6 and 95% CI for y-intercept, −47 to 49). The differences between the slopes and intercepts of the three equations are not statistically significant. 
×
Fig. 4. (  A  ) Distribution of percent prediction error in estimating intrathoracic blood volume (ITBV) in group C using  equation 7. (  B  ) Distribution of percent error in estimating ITBVefor group S showing the significant increase in error at 30 and 90 min (  open triangles  = time 0;  shaded triangles  = 30 min;  shaded squares  = 90 min;  open squares  = 150 min). (  C  ) Percent error in group S at 30 and 90 min using  equation 7compared with “corrected errors” when ITBVewas estimated using  equation 9(  shaded triangles  = errors at 30 min;  dotted triangles  = corrected errors at 30 min;  shaded squares  = errors at 90 min;  dotted squares  = corrected errors at 90 min). Percent prediction error refers to [(ITBVm− ITBVe)/ITBVm]× 100. 
Fig. 4. (  A  ) Distribution of percent prediction error in estimating intrathoracic blood volume (ITBV) in group C using  equation 7. (  B  ) Distribution of percent error in estimating ITBVefor group S showing the significant increase in error at 30 and 90 min (  open triangles  = time 0;  shaded triangles  = 30 min;  shaded squares  = 90 min;  open squares  = 150 min). (  C  ) Percent error in group S at 30 and 90 min using  equation 7compared with “corrected errors” when ITBVewas estimated using  equation 9(  shaded triangles  = errors at 30 min;  dotted triangles  = corrected errors at 30 min;  shaded squares  = errors at 90 min;  dotted squares  = corrected errors at 90 min). Percent prediction error refers to [(ITBVm− ITBVe)/ITBVm]× 100. 
Fig. 4. (  A  ) Distribution of percent prediction error in estimating intrathoracic blood volume (ITBV) in group C using  equation 7. (  B  ) Distribution of percent error in estimating ITBVefor group S showing the significant increase in error at 30 and 90 min (  open triangles  = time 0;  shaded triangles  = 30 min;  shaded squares  = 90 min;  open squares  = 150 min). (  C  ) Percent error in group S at 30 and 90 min using  equation 7compared with “corrected errors” when ITBVewas estimated using  equation 9(  shaded triangles  = errors at 30 min;  dotted triangles  = corrected errors at 30 min;  shaded squares  = errors at 90 min;  dotted squares  = corrected errors at 90 min). Percent prediction error refers to [(ITBVm− ITBVe)/ITBVm]× 100. 
×
Table 1. Hemodynamic Variables for Control and Shock Groups 
Image not available
Table 1. Hemodynamic Variables for Control and Shock Groups 
×