Previous studies found contradictory results regarding the question whether mixed venous oxygen saturation (Svo2) and central venous oxygen saturation (Scvo2) are equivalent. The inconsistency of study results may result from different study designs and different, partly questionable, statistical approaches.
The authors performed a prospective clinical trial comparing individual oxygen saturation values as well as the trend of values in blood from the superior vena cava (Scvo2), the right atrium (Srao2), and the pulmonary artery (Svo2) during varying hemodynamic situations. The subjects were 70 patients scheduled to undergo elective neurosurgical operations in the sitting position. Oxygen saturation was measured photospectrometrically in blood samples simultaneously taken at four different time points during supine and sitting positions. Statistical analysis was performed following the recommendations of Bland and Altman.
Five hundred two comparative sets of measurements were obtained. Ninety-five percent limits of agreement ranging from +/-6.83 to +/-9.30% for single values were interpreted as clinically unacceptable. In contrast, correlations between changes of Svo2 and Scvo2 as well as of Svo2 and Srao2 were interpreted as clinically acceptable (R > or = 0.755, Pearson correlation coefficient; P < or = 0.0001).
In this sample of patients, exact numerical values of Scvo2 and Srao2 are not equivalent to those of Svo2 in varying hemodynamic conditions. However, for clinical purposes, the trend of Scvo2 may be substituted for the trend of Svo2. In addition, previous studies investigating the agreement between Svo2 and Scvo2 were found to be lacking in their chosen statistical approaches.
MIXED venous oxygen saturation (Svo2) monitoring is used as a surrogate for the balance between systemic oxygen delivery and consumption during the treatment of critically ill patients.1However, measurement of Svo2requires placement of a pulmonary artery (PA) catheter with a risk-versus -benefit relation that is still a matter of controversy.2–4On the other hand, a central venous (CV) catheter is routinely inserted in critically ill patients for monitoring of CV pressure and administration of catecholamines and parenteral nutrition. Therefore, measurement of CV oxygen saturation (Scvo2) seems to be an attractive alternative to monitoring of Svo2because it can be performed more easily, is less risky, and is less costly.
In a recently published guideline, Svo2and Scvo2were declared as equivalent for the management of severe sepsis,5although previous studies found contradictory results.6–25It should be noted that the statistical approaches used in most of these investigations (e.g. , correlation and regression analysis7,8,12–14,16) are questionable regarding the statistical evaluation of the agreement of two different methods.26,27Furthermore, one has to differentiate between the analysis of single data points and the analysis of the change in oxygen saturation values over time (trend).
The aim of our study was to compare Svo2versus Scvo2in various hemodynamic conditions. Because some previous studies only evaluated the agreement between Svo2and right atrial oxygen saturation (Srao2),10,17,18we also compared Svo2versus Srao2. The sitting position, which is used to perform neurosurgical operations in the cerebellum and the cerebellopontine angle, is known to induce hemodynamic changes compared with the supine position.28–30Therefore, we prospectively studied the oxygen saturation of blood samples simultaneously taken during both the supine and the sitting position from the superior vena cava (SVC), right atrium (RA), and PA.
Materials and Methods
After obtaining approval from the Institutional Review Board at the University of Cologne (Cologne, Germany) and informed consent from each participant, we studied 70 patients scheduled to undergo a neurosurgical procedure in the sitting position.
Anesthesia Protocol and Catheter Placement
Anesthesia was induced and maintained with intravenous fentanyl and midazolam. Monitoring included electrocardiography, intravascular blood pressure measurement, pulse oximetry, CV pressure monitoring, urine output monitoring, precordial Doppler ultrasound for detection of venous air embolism, and arterial blood gas analysis.
Multiorificed CV catheters (Vygon GmbH, Aachen, Germany) were inserted via the right subclavian vein, and PA catheters (OptiQ®; Abbott, Chicago, IL) were inserted via the right internal jugular vein using a standard aseptic technique. Intravascular electrocardiography (Alphacard®; Sterimed, Puettlingen, Germany) was used to confirm the placement of the CV catheters in the lower SVC directly above the caval–atrial junction. The PA catheters were positioned following standard procedures. When the wedge position was obtained, the catheters were withdrawn to place the proximal injectate port in the RA, which was confirmed by intravascular electrocardiography.
Blood Gas Analysis and Determination of Hemodynamic Variables
Blood samples were drawn simultaneously from the PA, RA, and SVC at four different time points: (1) in the supine position after induction of anesthesia (T1); (2) twice during the neurosurgical operation in the sitting position (T2 and T3); and (3) in the supine position after the neurosurgical procedure was finished (T4). A standard volume of 1.5 ml blood was obtained from each site (QS 50®; Radiometer, Copenhagen, Denmark) and cooled in ice water after withdrawal of dead space blood and flushing fluid. Three consecutive oxygen saturations per blood sample were determined photospectrometrically immediately after each series of blood samples (OSM 3 Hemoximeter; Radiometer). The average of these three measurements was calculated for each site and each time point and used for further statistical analysis. Immediately after a series of blood samples was drawn, mean arterial blood pressure (MAP), heart rate, and cardiac output (CO) were recorded. CO was determined by bolus thermodilution technique using the PA catheter connected to a CO computer (Q-Vue®; Abbott). To quantify the repeatability of the standard oxygen saturation measurement method, in 17 patients, an additional five replicate blood samples were drawn immediately one after the other from the PA.
Based on unpublished data from our institution in a similar neurosurgical population, a power analysis was performed before the study, and it indicated that a sample size of 70 patients was necessary (α= 0.05, power = 0.80) to detect a difference of ±2% between Svo2and Scvo2, which was defined as the smallest clinically relevant difference. The anticipated SD was set at ±5%.
Each of the resulting five Svo2values per patient were subtracted from each other, resulting in n = 17 · 10 individual differences. In accordance with the recommendations of Bland and Altman,27a one-way analysis of variance with subject as the factor was calculated for these 170 individual differences. The within-subject SD sw was estimated as the square root of the residual mean square. Furthermore, a 95% repeatability coefficient was calculated as 1.96√2sw (= 2.77sw ), which is compared to the 95% limits of agreement between the different methods (see next section).
Statistical Analysis of Individual Values
The systematic error (bias) and the variability (SD of the bias) for Svo2versus Scvo2and Svo2versus Srao2were calculated. Bias was expressed as the mean of the differences of the individual values. The Student t test was used to determine whether the mean differences were significantly different from zero, and Pearson correlation coefficients between Svo2and Scvo2and between Svo2and Srao2were determined. Furthermore, 95% limits of agreement were calculated as bias ± 1.96 SD. The differences between individual values were plotted against their means for each time point (Bland and Altman plot).26
Statistical Analysis of Changes in Oxygen Saturation
For evaluating the agreement between the different sites of oxygen saturation measurement (Svo2vs. Scvo2and Svo2vs. Srao2) regarding the trend of values, a Pearson correlation coefficient was calculated for differences of oxygen saturation between sequential time points (T1–T2, T2–T3, T3–T4).
To quantify changes in hemodynamic variables, corresponding to the oxygen saturation measurements, differences between sequential time points were calculated for MAP, heart rate, and CO. Values are presented as mean ± SD unless otherwise stated. For all statistical procedures, a P value less than 0.05 was considered significant. The Statistical Package for the Social Sciences (SPSS, release 11.0; SPSS Inc., Chicago, IL) was used for all calculations.
Thirty-two male and 38 female patients with an average height of 170.0 ± 7.9 cm (range, 147–190 cm), weight of 76.1 ± 9.0 kg (40–120 kg), and age of 53.1 ± 8.8 yr (17–78 yr) were enrolled in the study. Five patients had an American Society of Anesthesiologists (ASA) physical status of I, 38 had an ASA physical status of II, 24 had an ASA physical status of III, and 3 had an ASA physical status of IV. Mean changes of hemodynamic variables between time points were as follows: MAP, 14.1 ± 9.6% (range, 2.4–42%); heart rate, 16.1 ± 16.0% (0–69.2%); and CO, 16.4 ± 14.3% (0–70%).
Evaluation of the agreement between different sites of oxygen saturation measurement requires knowledge about the repeatability of the oxygen saturation–measurement method itself.26Oxygen saturation values from 17 · 5 replicate blood samples revealed a within-subject SD sw of 1.05% and a repeatability coefficient of 2.91% (fig. 1).
Regarding the comparison of different sampling sites, ideally 2 · 280 comparative pairs of measurements from 70 patients at four different time points could be made in our study. However, because of technical (incorrect position of catheter) or organizational reasons (duration of operation), 256 (Svo2vs. Scvo2) and 246 (Svo2vs. Srao2) comparative sets of measurements, respectively, were obtained (table 1). The mean Svo2was larger than the mean Scvo2and the mean Srao2at all time points (T1–T4). Changing from the supine position to the sitting position (from T1 to T2) resulted in an increase in mean differences (bias) in oxygen saturation (Svo2vs. Scvo2and Svo2vs. Srao2), whereas changing from the sitting position to the supine position (from T3 to T4) led to a decrease in bias (Svo2vs. Scvo2and Svo2vs. Srao2) (table 1). Only once did the bias in oxygen saturations exceed 2% (Svo2vs. Scvo2; bias at T3 = 2.79%), which was defined as a clinically relevant threshold (table 1). Correlation coefficients ranged from 0.688 (Svo2vs. Scvo2; T4) to 0.851 (Svo2vs. Srao2; T2) (table 1).
More important are the distributions of differences of individual values from various sites of measurement. Figure 2depicts the Bland and Altman plot for Svo2versus Scvo2. Ninety-five percent limits of agreement for Svo2versus Srao2are calculated as ±7.12% (T1), ±6.83% (T2), ±7.32% (T3), and ±6.58% (T4). Differences in oxygen saturations of 10% or greater were observed in individual patients at all time points T1–T4 for comparisons Svo2versus Srao2and Svo2versus Scvo2.
Changes in oxygen saturation over time are presented in figure 3. Even when individual values were different, changes in Scvo2and Srao2paralleled changes in Svo2quantitatively, demonstrated by correlation coefficients of R = 0.75 (Svo2vs. Scvo2) and R = 0.82 (Svo2vs. Srao2), respectively (table 2). Considering more distinct changes in Svo2(> 5% and > 10%), correlation coefficients increased (table 2). All values of R are highly significant (P < 0.0001).
In the current study, the 95% limits of agreement between the standard Svo2method and both the Scvo2method and the Srao2method were large. In fact, some individual measurements of oxygen saturation of CV blood and RA blood differed more than 10% from corresponding mixed venous blood values. Therefore, an enormous variability was found between absolute values of Svo2and Scvo2and of Svo2and Srao2, respectively, suggesting that individual values of Scvo2and Srao2cannot substitute true Svo2values. However, the trend in Scvo2values as well as the trend in Srao2values demonstrated a good correlation with the trend in Svo2values.
The relation between Scvo2and Svo2has been examined in numerous studies with controversial conclusions.6–25The wide range of conclusions might be the result of different study designs and of different statistical approaches (table 3). Some studies used animal models,12,14,17whereas other studies examined healthy volunteers21or patients with myocardial infarction.24The number of subjects ranged from 79to 61,7and the number of comparative pairs of measurements ranged from 276to 580.9In most studies, an unequal number of blood samples per subject was drawn,8–14,16,21,25potentially resulting in an imbalanced statistical weight of subjects with exceedingly good or bad correspondences of Scvo2and Svo2. In addition, in some studies, blood samples were not drawn simultaneously, but sequentially during the advancement of the PA catheter,6–8,16,21possibly causing arrhythmia31and thus significantly altering hemodynamic variables, which may result in different oxygen saturations at different sites of measurement.
Neither a power analysis nor an analysis of method repeatability was presented in previous investigations. A power analysis has to be calculated on the basis of a difference between Svo2and Scvo2that is defined as clinically relevant before the study. Furthermore, this Δ (Svo2− Scvo2) value serves as an a priori standard, which is necessary to discuss the variability found in the actual study data. Evaluation of the repeatability of single measurements is a very important issue because the repeatability of methods limits the amount of agreement that is possible.27Therefore, repeatability represents a baseline (within-method variability) to which the between-method variability can be compared.
Most studies present the bias6–8,10,11,13,15–20as well as correlation coefficients between the measurements.7–20,25However, calculating bias and correlation coefficients for individual values is not enough for evaluating the agreement of two different methods.26,27Despite large differences between individual values, the bias (i.e. , mean of the differences between individual values) might be zero, and the correlation coefficient measures proportionality, not agreement. In 1986, Bland and Altman26recommended the calculation of 95% limits of agreement of individual values for method comparison studies. However, only 5 of 10 investigations studying the equivalence of Svo2and Scvo2have presented these limits of agreement since then.9–11,14,18Furthermore, only 8 of 17 studies analyzed the agreement of the trend of oxygen saturations measured at different sites in addition to the agreement of absolute values.8–13,15,17
The wide range of 95% limits of agreement regarding individual values of Svo2and Scvo2found in our study indicates a large between-method variability. Because the 95% repeatability coefficient was 2.91% in our study, this considerable lack of agreement between the methods is not solely explained by a lack of repeatability. Our finding parallels the results of previous studies, demonstrating an enormous variability of absolute values.6,8–13,16,18In contrast, some studies found that individual Scvo2values can adequately replace Svo2values,7,15,17,20,24,25but these studies do not present a statistical analysis that includes more than the evaluation of bias and correlation coefficients. In addition, none of these studies performed a power analysis before the investigation. Therefore, it remains unclear whether a number of 4415,25or 7620comparative measurements is enough to detect statistically significant differences. Schou et al. 14concluded from their data that Svo2and Scvo2are interchangeable, although the 95% limits of agreement plotted in their figures equaled or exceeded ±10%. Because of these controversial results regarding absolute values, there has been a considerable debate on the question of the clinical utility of Scvo2.1,32
Clinical decisions are rarely based on single measurements but always reflect various variables as well as the trend of these variables. In the early stage of a disease, Scvo2values are often found to be less than 50%, with Svo2values even lower.1Consequently, these low Scvo2values, although they do not exactly equal Svo2values, may serve as a representative variable guiding therapeutical interventions. Furthermore, not the individual value but the trend of Scvo2may detect an imbalance of oxygen delivery and oxygen consumption. Our study reveals a good correlation between the trends of Svo2and Scvo2, which is in accordance with previous studies.11–13,15Scheinman et al. 13found a poor correlation between absolute values in patients with severe heart failure or shock but a better correlation between changes of Svo2and Scvo2. In an animal model, Reinhart et al. 12demonstrated a close tracking of the oxygen saturations continuously measured in the PA and the SVC across a wide range of hemodynamic conditions. In a recent study, the same group could confirm these findings in critically ill patients.11Tahvanainen et al. 15found a significant correlation between PA blood samples and both SVC and RA blood samples during subsequent changes of oxygen saturation in critically ill patients. These data suggest that Scvo2is equivalent to Svo2in the course of clinical decisions as long as absolute values are not required. This is supported by our data, because the degree of correlation between the trend of Scvo2and Svo2is better with larger changes in oxygen saturation. In a recent study, Rivers et al. 33significantly reduced mortality and organ dysfunction in patients with severe sepsis or septic shock using an early goal-directed therapy approach. The early goal-directed algorithm of the treatment of the first 6 h included the continuous measurement of Scvo2defining an Scvo2value greater than 70% as an endpoint of therapy.33Although some variables of the algorithm, including Scvo2, have been criticized,34monitoring Scvo2has recently been declared as appropriate for the management of severe sepsis.5
In addition, in our study, a comparison between Svo2and Srao2could be performed using the proximal port of the PA catheter for withdrawal of blood samples from the RA. However, 95% limits of agreement were too large to accept individual Srao2values as equivalent to Svo2. The trend of Srao2correlated better with Svo2than those of Scvo2. However, because of potential dangerous complications (e.g. , arrhythmia, perforation), placing the catheter tip in the RA is not recommended35or at least debatable.36
Limitations of the Study
Monitoring Svo2is used as a clinical marker of systemic oxygen utilization in critically ill patients.1During sepsis, a redistribution of blood flow associated with a proportionally greater reduction in splenic, renal, and mesenteric blood flow may occur,1,32and in heart failure or circulatory shock, blood flow is relatively increased in cerebral and coronary circulation,8,13thus altering the relation between Svo2and Scvo2, with Svo2generally lower than Scvo2. However, our study was performed in patients scheduled to undergo elective neurosurgical operations in the sitting position during general anesthesia, resulting in a mean Svo2higher than Scvo2. This study design was chosen because, in our neuroanesthesia department, these patients are routinely monitored with both a PA catheter and a CV catheter (for aspiration of entrained air), whereas critically ill patients are rarely provided with both catheters. Second, this setting, in contrast to the clinical situation of critically ill patients, allows withdrawal of blood samples at different time points that are clearly defined. Third, positioning the patient from the supine position to the sitting position and vice versa was reported to induce a clear change of hemodynamic variables28,29due to a change in blood volume distribution from the intrathoracic to the extrathoracic compartment,30which is demonstrated for the current study. A mean change of 14% in MAP, with ΔMAP greater than 40% in some patients, and a mean difference in CO of 16%, with a maximum ΔCO of 70%, were detected. This might partly mimic the situation of patients in hypovolemic shock and resulted in an increase in bias between Svo2and Scvo2after changing the position from supine to sitting and vice versa in a decrease in bias after changing the position from sitting to supine. Anesthesia was induced and maintained with intravenous fentanyl and midazolam in all patients. Because both drugs are often used as a sedative or an analgesic on the intensive care unit,37,38the anesthesia regimen used in our study is not significantly different from sedation regimens used in critically ill patients. Furthermore, fentanyl and midazolam are reported to produce only modest hemodynamic effects.39,40Therefore, it is unlikely that our findings are significantly affected by the specific anesthesia regimen.
In summary, our study demonstrates that despite some large differences between absolute values, in patients with varying hemodynamic situations, the trend in Scvo2may be used as a surrogate variable for the trend in Svo2.
The authors thank Gernot Wassmer, Ph.D. (Assistant Professor, Institute of Medical Statistics and Epidemiology, University of Cologne, Cologne, Germany), for statistical advice.