Cerebral oximetry is a noninvasive, optical technology that integrates frontal cortex blood hemoglobin-oxygen saturation. The technology evolved from efforts to measure the state of oxygen metabolism in tissues by assessing the near-infrared absorption of cytochrome C oxidase, a function of oxygen partial pressure in the mitochondrion.1,2 The assessment of cytrochrome oxygenation state in complex tissues in vivo is challenging because the near-infrared absorbance of hemoglobin overlaps with that of the cytochromes. Furthermore, hemoglobin is present in most tissue regions of interest at far higher concentrations than cytochrome C. It is therefore not surprising that near-infrared spectroscopy of cytochromes is subject to substantial errors due to difficulty assigning the correct computational algorithm to account for oxy- and reduced hemoglobin in tissue.3,4 In contrast, oxygen use with insufficient oxygen delivery will lead to an increase in reduced hemoglobin, and since hemoglobin chromophores dominate the near-infrared absorption spectrum, the estimation of hemoglobin-oxygen saturation should be less susceptible to these issues and errors. Similar to pulse oximetry, it should theoretically be possible to calibrate a cerebral oximeter’s reading (regional saturation, rSO2or the cerebral oximeter’s estimate of regional saturation, ScO2) to the ratio of oxy- and deoxyhemoglobin in the vasculature. An in vivo calibration can be accomplished with direct measurement of cerebral venous and arterial saturation in blood samples. This calibration, incorporated into the calculation algorithm that determines the displayed value of ScO2, should theoretically reduce the interfering effects of different hemoglobin concentrations and variations in tissue light transmission. This calibration approach assumes that all individuals have the same ratio of cerebral mixed venous and arterial blood within the tissue where saturation is measured.
In recent years, the clinical use of cerebral oximetry has expanded considerably and there are at least 4 companies selling cerebral oximeters in North America. Clinical interest in the devices is substantial, in part because of the hope that brain oxygenation state assessed by cerebral oximeters might predict important long-term outcomes and complications. Among the outcomes predicted by low cerebral oximeter readings include neuropsychiatric impairment,5 stroke in cardiac surgery patients,6 organ dysfunction and mortality,7,8 and length of hospital stay.9 Similarly, cerebrovascular insufficiency as assessed by cerebral oximetry10 may be a marker for poor clinical outcome in cardiac surgery patients. However, it remains unclear whether cerebral oximetry serves as a reliable clinical monitor in carotid endarterectomy,11 head injury,12 and in the pediatric population.13 Contamination of signals from extracranial blood in the sampled area also complicates the clinical utility of this technology.14,15
The comparative accuracy of cerebral oximeters has not been independently evaluated in detail since a study by Kim et al.16 more than a decade ago. The Kim et al.16 study compared the rSO2 estimated by the INVOS 4100A cerebral oximeter (Somanetics Corp, Troy, MI) with that of jugular venous blood, and not the weighted average of arterial and venous blood that is the stated basis for cerebral rSO2 determination in currently available instruments. There are currently no Food and Drug Administration (FDA) standards for accuracy of cerebral oximeters as with pulse oximeters. This technology differs from pulse oximetry in the important respect that it cannot provide an absolute index of saturation. This explains our clinical observation that even in well-oxygenated and normal subjects or patients, repeated readings in the same patient vary substantially from one another, and that during hypoxia the differences persist and may expand. Importantly, the causes of the variation in the difference between repeated readings between subjects are not apparent. We believe that identifying the contributing factors to interindividual variability in cerebral rSO2 could improve cerebral oximeter technology. Accordingly, the purpose of this study was to determine and quantify the factors that contribute to bias between cerebral oximeter reading and the weighted average of cerebral mixed venous and arterial blood saturation in healthy subjects during isocapnic hypoxia.
The University of California at San Francisco (UCSF) Committee on Human Research approved the study, and all subjects gave informed written consent. The pool of subjects were healthy nonsmoking men and women, from 20 to 49 years of age, willing to volunteer for the study for a nominal payment. The selected group of subjects was gender and ethnically balanced, following the U.S. FDA requirements for standard studies of pulse oximeter accuracy.a With an enrollment target of 25 subjects, we enrolled 23 healthy adult subjects, 14 men and 9 women, with a range of skin pigmentation. This enrollment target was set because studies to evaluate pulse oximeter accuracy typically involve 10 to 12 subjects per U.S. FDA guidelines, and we expected cerebral oximeter errors to be larger than for finger pulse oximeters. Skin pigment was classified as light, intermediate, or dark, as in a previous publication concerning the effect of skin pigment of pulse oximeter accuracy.17 Pregnancy in women was excluded by medical history. The data were analyzed after completion of the study.
A 22-gauge radial arterial cannula was placed using 1% lidocaine local anesthesia in either wrist of each subject for arterial blood sampling and continuous measurements of arterial blood pressure. The volunteers were placed in the Trendelenburg position and a 20-gauge 5-inch catheter (Arrow International, Reading, PA) was inserted retrograde into the right internal jugular vein, using sterile technique, local anesthesia with 1% lidocaine, and ultrasound guidance (Site Rite, Dymax, Pittsburgh, PA) to identify the vein. The catheter was advanced the entire 5 inches or until the subject noted a sensation in the jaw or ear area, indicating that the tip of the catheter was in the jugular bulb. After catheter insertion, ultrasound was used to verify the location of the catheter tip in the jugular bulb and further adjustment done as needed.
We evaluated the accuracy performance of 5 commercially available cerebral oximeters in routine, worldwide use: the EQUANOX® 7600 in 3- and 4-wavelength versions (Nonin Medical, Plymouth, MN), FORE-SIGHT® (Casmed, Branford, CT), INVOS® 5100C (Covidien, Boulder, CO), and the NIRO-200NX® (Hamamatsu Photonics, Hamamatsu City, Japan). The oximeter sensors, 2 at a time per subject, were randomized to the right or left side of the forehead and covered with dark cloth and aluminum foil to shield them from ambient light. Each subject had all 5 cerebral oximeter sensors placed, in random order, during 3 runs of hypoxia. Its sensor was adjusted if an oximeter did not display a saturation value. The output of each oximeter was recorded both by serial datastream acquisition and by manually recording the display reading for rSO2 from each device, as a backup. The serial data output was used for analysis.
To measure cerebral oximeter performance during hypoxia, the fraction of inspired oxygen (FIO2) was stepwise changed to produce stable oxygen saturation (SaO2) levels between 70% and 100%, while the end-tidal CO2 was continually monitored on a computer screen and maintained constant by adding CO2 to the breathing gas as needed. We used a semi-open rebreathing system that allowed very stable plateaus of oxygen saturation and CO2 partial pressure. Blood samples from the jugular bulb and radial artery were simultaneously collected. Before each blood sample, “dead space” blood was removed from the catheters and discarded. Arterial and venous blood samples were analyzed with a multiwavelength optical blood analyzer (OSM-3, Radiometer Medical A/S, Copenhagen, Denmark) to determine oxygen saturation (SaO2 and SvO2).
The rate of blood draw from the jugular catheter was 1 mL over 30 seconds, to reduce “reflux” of nonjugular bulb blood into the sample.18 The ratio of saturation in arterial and venous blood claimed by each instruments manufacturer was used to calculate the “reference saturation.” For INVOS, the weighting ratio was 25/75 arterial/venous and for the others 30/70. The accuracy of the instruments could then be calculated from this reference saturation and the reading from the cerebral oximeter.
At the beginning of each study, 1 arterial and 1 venous blood sample was drawn from each subject while they breathed room air. Hypoxemia was then induced to 5 to 6 different targeted stable SaO2 levels (between 70% and 100%, based on end-tidal gas analysis) by having subjects breathe mixtures of nitrogen, air, and CO2 according to an established protocol, previously detailed.17 Each SaO2 level was maintained at a steady state for at least 60 seconds, allowing oximeter readings to stabilize, at which point 2 simultaneous arterial and 2 jugular bulb venous blood samples were obtained, 30 seconds apart during the steady state. The target saturation values below room air, and nominal PaO2 corresponding to each were as follows: 92%, 63 mm Hg; 87%, 53 mm Hg; 82%, 46 mm Hg; 77%, 42 mm Hg; and 70%, 37 mm Hg.
Hypotheses and Statistics
We did not have preliminary data to perform a power analysis. Twenty-five subjects were chosen to double the 12 subjects typically required for U.S. FDA evaluation of pulse oximeter accuracy.
The following hypotheses and statistical approaches were used in this study:
Cerebral oximeters have a bias that is not affected by the reference (weighted) saturation range. The weighting is specified by each manufacturer (75%–25% cerebral mixed venous to arterial for INVOS, 70%–30% for the others). The following standard descriptive statistics for performance were calculated: the bias for each cerebral oximeter reading (oximeter reading-weighted saturation), the mean bias of all readings for each instrument, a regression line for a plot of bias versus weighted saturation (the “reference”), precision (standard deviation of the bias), limits of agreement and the root mean square error (Arms). The Arms was calculated for different ranges of reference oxygen saturation (weighted value of SaO2 and SvO2) as the square root of the arithmetic mean of the individual errors. This statistic is used as a standard measure of performance expressing the variability of the errors in measurements, including pulse oximeter errors.
The linear regression of the bias plots was analyzed according to the method of repeated measures by entering the subject number as a random effect in the model. The between-subject variation (sum of squares), within-subject variation (variation due to differences in the reference saturation), and error (residual) were determined for each cerebral oximeter. Limits of agreement for the bias plots were calculated with the methods of Bland and Altman19 with adjustments for multiple measurements for each individual. In subsets within ranges of the reference saturation without sufficient numbers of data points, limits of agreement were calculated as ± 1.96·SD. Components of the model (between subject, SaO2/SvO2 reference, and residual error) were determined.
Data were also analyzed by comparing the mean bias within ranges of reference saturation (30%–50%, 50%–70%, and 70%–90%). The Shapiro-Wilk test was used to test the normality of the distribution of individual bias values. This statistic revealed that distribution of bias was slightly skewed in all 5 instruments (all P < 0.01). However, the differences in mean bias for defined ranges of reference saturation (SaO2/SvO2) were sufficiently large that we were justified in using analysis of variance for repeated measures without any data transformation. Tukey-Kramer honest significant difference was used for multiple comparisons.
Changes in the mean bias (difference between cerebral oximeter reading and weighted measurement of cerebral mixed venous and arterial blood) of cerebral oximeters during hypoxia are influenced by the assumption that cerebral oximeters detect the saturation of a fixed ratio of cerebral mixed venous and arterial blood. This means that bias calculated based on the arterial-venous saturation difference (SaO2 – SvO2) differs from the bias calculated based on the weighted difference of measured saturations. This hypothesis was tested by determining the significance of slopes of plots of bias plots calculated with weighted saturation and with bias calculated by the arterial to venous (A − V) difference. If the slope was significant in one instance and not in the other, this would support the hypothesis.
An early inspection of the data revealed that there was large reading variation (i.e., bias) among different subjects, but that within subjects the bias was distributed much more tightly. For example, one reason why an individual’s baseline reading of cerebral saturation may be different from another’s is that he/she has a different ratio of venous and arterial blood in the sensor field; this condition creates an “offset” in bias compared with the mean bias of all subjects. This effect was described, not by a particular statistical test, but by partitioning the percentage of variation in bias values caused by between-subjects variation and the percentage of variation in bias caused by changes in oxygenation (changing arterial saturation between 100% and 70%). This was calculated using the JMP statistical package (SAS Institute, Cary, NC), using the same repeated-measures model in the regression analysis (Fig. 1). The output supplies the sum of squares for the components in the model: subjects, the X variable (the reference saturation), and the residual error.
Hypotheses 3 and 4
Cerebral oximeter mean bias varies by skin color and gender. The effect of gender and skin color on bias was examined with a multivariate analysis. To construct a multivariate analysis examining the impact of skin color and gender on bias, we started with the repeated-measures model described above, analyzing bias versus the weighted (reference) saturation (SaO2/SvO2). To this, we added the other variables, i.e., skin pigment or gender. The interaction term was the reference multiplied by either variable. Although the data were not strictly normally distributed as assessed by the Shapiro-Wilk test, we used a linear model for the multivariate analysis because the method appeared sufficient to identify the effects of skin or gender on the bias. For example, log transformation of the bias did not improve the statistical significance of the effect of skin pigment or gender in the model. The model also did not account for unequal covariances (“sphericity”). These inadequacies could produce a type 2 error, but would not have a significant likelihood of producing a type 1 error. The Tukey-Kramer honest significant difference method was used to assess multiple comparisons among the 3 different categories of skin color (light, intermediate, dark). Because of the assumptions related to bias distribution, we set significance at a P < 0.001.
Data are reported as mean ± SD or mean (95% confidence interval [CI]) as indicated. Data were analyzed with JMP 10.0 (SAS Institute) or using custom-generated functions in Excel 14.2.2 (Microsoft, Redmond, WA). For all statistical tests, P < 0.05 was considered significant.
Table 1 is a demographic summary of the 23 subjects enrolled in the study.
Bias of Cerebral Oximeter Readings Compared With Weighted Blood Saturation Measurements During Hypoxia
Five hundred forty-two comparisons between pairs of arterial and venous blood samples and oximeter readings were analyzed in the 23 subjects. For each paired blood draw, the weighted value of arterial and venous saturation was used to compute a bias by comparing this weighted saturation value with the instrument reading. Table 2 contains summaries of this bias in terms of the mean bias, standard deviation of the bias, SE, and root mean square (Arms) error values for specific ranges of calculated weighted arterial and venous saturations. The pooled Arms error, tallied from all 5 instruments, was 9.1%. Over the entire range of oxygenation studied, the mean bias ± SD (precision) and Arms errors were: FORE-SIGHT 1.76 ± 3.92 and 4.28; INVOS 0.05 ± 9.72 and 9.69; NIRO-200NX −1.13 ± 9.64 and 9.68; EQUANOX 3-wavelength 2.48 ± 8.12 and 8.47; EQUANOX 4-wavelength 2.84 ± 6.27 and 6.86.
Plots of cerebral oximeter bias versus the hemoximeter-measured weighted value for rSO2 are presented in Figure 1. The FORE-SIGHT, NIRO-200NX, and EQUANOX 3-wavelength had significantly greater bias at lower SaO2, i.e., the slope of the bias plots for these 3 instruments was negative (Fig. 1, P < 0.0001 for significance of slope in all 3 cases). The difference in mean bias between the high and low ends of the saturation was relatively large (Table 2), meaning that we have rejected hypothesis 1 for these instruments. For example, for the FORE-SIGHT, there was a difference in bias of about 10% saturation between weighted blood saturations of 85% and 50%; for the NIRO-200NX, the bias difference over this range was 15%; and for the EQUANOX 7600 3-wavelength instrument, the bias difference was 11%.
To determine whether increasing positive bias of the cerebral oximeters with hypoxemia is related to the assumption of venous/arterial blood volume weighting, we plotted bias against the difference between arterial and cerebral mixed venous saturations (i.e., nonweighted saturation) in Figure 2. In these plots, the slopes of bias in the EQUANOX 3-wavelength (0.12; CI, 0.02–0.22), FORE-SIGHT (0.06; CI, −0.02 to 0.14), and NIRO-200NX (−0.14; CI, −0.27 to −0.013) instruments do not depend on the nonweighted saturation, i.e., the slopes are not statistically different from zero. This suggests that factors intrinsic to the difference in venous and arterial saturation (e.g., changes in cerebral blood flow or cerebral metabolism) may contribute to the differences in bias between the high and low end of the saturation spectrum in these 3 instruments (hypothesis 2).
Analysis of Interindividual Differences in Cerebral Oximeter Bias
Substantial differences in cerebral oximeter readings among individuals were also observed. Between-subject differences account for 36% to 87% of the variation in data among the 5 cerebral oximeters. This was evident when inspecting the bias values for individual subjects in the plots in Figure 1, A to E. For each instrument, regression lines of bias plots for individual subjects appeared to have negative slopes with respect to the weighted A − V difference, with either a positive or negative bias offset. The negative slope of the bias plots (i.e., more positive bias with hypoxia) is based on increasing bias of individual subjects’ response at low oxygen levels and not an artifact of bias being grouped for individuals at the high and low end of the saturation ranges. These patterns can be seen in Figure 3 in which we have plotted regression lines for individual subject bias values for a randomly chosen instrument (random number generator), the Nonin EQUANOX 3-wavelength cerebral oximeter. The percentage of variation attributable to between-subject variability was, for EQUANOX 3-wavelength 76%, EQUANOX 4-wavelength 61%, FORE-SIGHT 37%, INVOS 87%, and NIRO 77%, with the remainder variation due to differences in reading because of changing saturation. In contract, in a recent study of standard pulse oximeter performance in our laboratory, the source of variability was nearly equally divided between changes in saturation and between-subjects effects (John Feiner, UCSF, 2012, unpublished data).
To understand some of the basis for the differences in bias among individuals, we performed a multivariate analysis to isolate several factors previously established to influence pulse oximeter bias. First, we examined the influence of skin pigmentation on cerebral oximeter bias (hypothesis 3), since we found that darkly pigmented skin causes a positive mean bias in pulse oximeter readings at low saturation.17,20 In Table 3 are shown the mean bias values according to type of cerebral oximeter and the skin pigmentation of the subject, and Figure 4 contains mean bias for 3 different ranges of weighted arterial-cerebral mixed venous saturation segregated by skin color groups. While it appeared that bias was quantitatively more negative in darkly pigmented subjects in all instruments, the repeated-measures analysis only showed a statistically significant difference in the FORE-SIGHT cerebral oximeter (Table 3, multivariate analysis of skin color by reference saturation, adjusted P < 0.001). The differences in bias between light and intermediate or darkly pigmented subjects indicated that darker skin tends to make this oximeter read lower than the weighted average of cerebral mixed venous and arterial blood, i.e., the bias values are generally more negative in groups of subjects with intermediate or darkly pigmented skin. Forty-eight percent of the subjects were of intermediate or dark skin pigmentation, and 52% were Caucasian. Even though the distribution of bias was not normal, and we did not account for unequal covariances, the degree of statistical significance (P < 0.001) for the interaction term (skin pigment times reference saturation) made a type 1 error (falsely concluding a significant effect of skin pigment on bias) unlikely in this case.
We also examined the factor of gender on bias (hypothesis 5) in cerebral oximeter readings (Table 4). In a multivariate analysis, we found that gender significantly affected the magnitude of the cerebral oximeter bias in the INVOS instrument (Fig. 5), setting P < 0.001. The interaction of bias and gender was also significant (P < 0.001) in a multivariate analysis for all instruments except the EQUANOX 4-wavelength (P = 0.007) and FORE-SIGHT (P = 0.92, Table 4).
Effects of Hypoxia on Arterial to Cerebral Mixed Venous Saturation
In Figure 6, we plotted the A − V saturation difference versus the arterial saturation for 6 representative subjects. The slope of this relationship was significant, P < 0.0001.
Cerebral Oximeter Performance
Five cerebral oximeters all detected decreases in cerebral blood oxyhemoglobin saturation during systemic hypoxemia in 23 healthy volunteer subjects. Based on the manufacturer-defined ratio of cerebral venous and cerebral arterial blood detected by each device, we calculated the bias between the cerebral oximeter reading and the weighted values of arterial and cerebral mixed venous saturation. As summarized in Table 2 and Figure 1, the FORE-SIGHT, NIRO 200-NX, and EQUANOX 3-wavelength instruments have a significantly positive mean bias during hypoxia. They indicate a cerebral saturation significantly higher than that based on simultaneous arterial and cerebral mixed venous measurements, whereas the other instruments have no significant mean bias. However, the variability in baseline readings among the 5 instruments was substantial, with significant between-subject and between-instrument variation. This indicates that unlike pulse oximeters, currently manufactured cerebral oximeters do not provide an “absolute” measurement of oxyhemoglobin saturation in the tissue region of interest, despite the theory that spatially resolved spectroscopy can determine a scaled tissue hemoglobin concentration and therefore the relative concentrations of oxy- and deoxyhemoglobin. The between-subject variability and dynamic error of readings makes it difficult to determine the absolute threshold of saturation that results in tissue damage. This is currently an important disadvantage of noninvasive optical brain oximetry. Until the factors that contribute to the variability are better understood, the utility of cerebral oximetry will remain limited. This is essentially the same conclusion made by Henson et al.21 in 1998 after a study of the INVOS 3100 cerebral oximeter (Somanetics Corp). Henson et al.21 found that while the INVOS 3100 was capable of detecting changes in oxygenation, the relationship between instrument reading and hypoxia exhibited a variety of slopes and intercepts between different individuals, not unlike the plot in Figure 3.
Causes of Cerebral Oximeter Bias
Manufacturers of cerebral oximeters have apparently used simultaneous measurements of arterial and cerebral mixed venous saturation to create algorithms for cerebral oximeter output, similar to our protocol. This process pools data from individuals with different ratios of arterial and venous blood but assumes a fixed ratio of 70/30 or 75/25 venous to arterial blood volume. The 70/30 ratio of venous to arterial blood is based on measurements of brain arterial and venous blood volumes with positron emission tomography.22,23 This assumption, and measurements of saturation in venous and arterial blood samples, can be used to calibrate an instrument to produce an output that is sensitive to desaturation (as was found, all 5 oximeters are good at detecting decreases in rSO2 during isocapnic hypoxia), but will not address the differences in arterial/venous mixtures among individuals. This could be the fundamental reason that the Arms of cerebral oximeters is so much greater than with pulse oximetry (9% vs 2%–3%). Unlike pulse oximeters, cerebral oximeters have no method similar to relating saturation to the ratio of light absorbance by pulsatile and nonpulsatile signals, so they cannot easily cancel out interfering light absorption. For cerebral oximeters, contributions to systematic bias, such as skin pigmentation or gender (Tables 3 and 4 and Figs. 4 and 5), will similarly not be corrected by simply averaging data from a broad population of subjects; this approach is appropriate for calibrating the average effects of desaturation but not for reducing bias among individuals. Manufacturers may also use other signal processing and interference correction algorithms to improve performance, but it is unclear how these relate to differences in performance among the 5 cerebral oximeters.
When we plotted the bias between ScO2 and Sa/vO2 against the difference in arterial and venous saturation, we found that a greater bias at low saturation was eliminated in the instruments showing this bias (Fig. 2). This suggests that differences in the ratio of venous to arterial blood volumes in different subjects accounts for at least some of the more positive bias at low saturation in these instruments, assuming that oxygen consumption of the brain remains the same and that the equation rSO2 = α(SaO2) + (1 − α)SvO2 applies where α is the fraction of arterial to venous blood volume in the brain tissue illuminated by the oximeter. Thus, changes in arterial and venous blood volume would produce changes in cerebral oximeter bias during maneuvers that change this ratio, including changes in cerebral blood flow produced by hypoxemia, hypercapnia or hypocapnia, changes in posture, venous outflow obstruction, etc. However, this was apparently not the case in the Kim et al.16 study where changes in CO2 did not change bias, although the bias plots in the Kim et al.24 study compare cerebral oximeter readings with only the average of jugular venous and radial arterial saturation and not the weighted average of venous and arterial as would be evaluated by the cerebral oximeter. Isocapnic hypoxia increases cerebral arterial blood flow in humans, probably causing increases in arterial blood volume relative to venous volume in the frontal cortex. Figure 6 shows that the A − V saturation difference decreased during hypoxia in all our subjects. For cerebral oxygen delivery to be constant, arterial blood flow must increase. This almost certainly means that arterial volume increases to some degree during hypoxia, violating the constant A − V volume assumption.
Using positron emission tomography techniques, and reviewing data from several types of experimental studies in humans and experimental animals, Ito et al.22,23 concluded that changes in CO2 produce changes in arterial blood volume and not venous volume. Although we maintained normocapnia throughout our studies, we expect that arterial blood volume would increase with the cerebral vasodilation produced by hypoxemia. Accordingly, the ratio of venous to arterial blood volume probably changes during the course of our experiments (unlike the case for hypo- and hypercapnia, we are unable to locate studies measuring relative cerebral venous and arterial blood volumes during isocapnic hypoxia). If this assumption is true, then the amount of arterial blood “seen” by the cerebral oximeter would increase during hypoxia, producing an overestimation of true rSO2. This type of error causes a positive bias in cerebral oximeter readings at low saturation, as was seen to varying degrees in the instruments tested. The effect was minimal in the 4-wavelength instrument from Nonin and the INVOS 5100C, but significant for the Casmed, Hamamatsu, and 3-wavelength Nonin instruments.
Clinical Implications of Cerebral Oximeter Performance
Cerebral oximetry tracks changes in saturation better than absolute saturation, and pulse hemoglobinometry tracks changes in hemoglobin better than absolute hemoglobin.25 Both methods have large interindividual differences. This is a major barrier to wider adoption of this and similar technologies.
The apparently significant influence of venous to arterial blood ratios on cerebral oximeter bias has implications for interpreting cerebral oximeter readings in clinical states that involve changes in the ratio of venous and arterial blood in the sensor field. These conditions include changes in intracranial pressure, venous outflow obstruction, use of vasoactive drugs and anesthetics, and alterations in body position. Studying how these alterations influence cerebral oximeter readings could lead to improvements in cerebral oximeter performance.
In practical terms, the interindividual differences in saturation are a major limitation of present cerebral oximetry technology because it is difficult to set a threshold for a dangerous lower level for ScO2. If readings among individuals vary by 20%, how can a threshold for concern be established? At present, the trend in values in an individual may be more useful and indicative of brain well-being than an absolute value, although the different slopes of individual subjects on the bias plots make this aspect of cerebral oximetry problematic as well.
Similar to previous studies involving finger pulse oximeters, skin pigmentation and gender were related to bias for some instruments.17,20 In the case of darkly pigmented skin, more negative bias was observed (although was statistically significant only for FORE-SIGHT); this is the opposite of the effect in pulse oximetry where darkly pigmented skin produced higher pulse oximeter readings at low saturation. Since the direction was the same in all the devices, given the few data points in darkly pigmented individuals, our data may not have been sufficiently powered to detect differences. Nonetheless, we speculate that more darkly pigmented skin is “seen” by the cerebral oximeter as more venous, biasing the reading toward lower saturation values. In the case of gender, all instruments except FORE-SIGHT read lower in women, across the spectrum of saturations. Whether this is related to lower hemoglobin, thinner cranial bone, or other factors in female subjects is not clear.
Because our study involved only a small number of very darkly pigmented subjects, our study has limited power to conclude that dark skin creates cerebral oximeter reading errors. Another limitation is that all subjects were studied in a 30° head-up position and this might cause bias compared with supine position. Another source of error may have been introduced by the position of different sensor types next to each other in the same subject (i.e., right and left sides). It is possible that even with shielding, light interference might occur, but since location and type of neighboring sensor were randomized, this should have had a minimal effect on our overall conclusions.
Five commercially available cerebral oximeters tracked changes in cerebral blood oxygenation during hypoxia in healthy volunteers. However, significant imprecision between weighted arterial-cerebral mixed venous saturation and cerebral oximeter readings were found, with an average RMS error of 9.1%. Differences in skin pigment, gender, and the assumed mixture of arterial and cerebral venous blood in the detection field likely contribute significantly to the bias and imprecision in readings from these devices.
Name: Philip E. Bickler, MD, PhD.
Contribution: This author designed and conducted the study and prepared the manuscript.
Attestation: Philip Bickler approved the final manuscript, attests to the integrity of the original data and the analysis reported in this manuscript, and is the archival author.
Name: John R. Feiner, MD.
Contribution: This author designed and conducted the study and prepared the manuscript.
Attestation: John Feiner approved the final manuscript and attests to the integrity of the original data and the analysis reported in this manuscript.
Name: Mark D. Rollins, MD, PhD.
Contribution: This author helped conduct the study and edited the manuscript.
Attestation: Mark Rollins approved the final manuscript and attests to the integrity of the original data and the analysis reported in this manuscript.
This manuscript was handled by: Dwayne R. Westenskow, PhD.
a U.S. Food and Drug Administration. Pulse Oximeters-Premarket Notifications Submissions [510(k)s]-Guidance for Industry and Food and Drug Administration Staff. Available at http://www.fda.gov/downloads/MedicalDevices/DeviceRegulationandGuidance/GuidanceDocuments/UCM081352.pdf. Accessed May 21, 2013.
1. Chance B. Spectrophotometry of intracellular respiratory pigments. Science. 1954;120:767–75
2. Jöbsis FF. Noninvasive, infrared monitoring of cerebral and myocardial oxygen sufficiency and circulatory parameters. Science. 1977;198:1264–7
3. Boas DA, Franceschini MA. Haemoglobin oxygen saturation as a biomarker: the problem and a solution. Philos Trans A Math Phys Eng Sci. 2011;369:4407–24
4. Devor A, Sakadžić S, Srinivasan VJ, Yaseen MA, Nizar K, Saisan PA, Tian P, Dale AM, Vinogradov SA, Franceschini MA, Boas DA. Frontiers in optical imaging of cerebral blood flow and metabolism. J Cereb Blood Flow Metab. 2012;32:1259–76
5. Yao FS, Tseng CC, Ho CY, Levin SK, Illner P. Cerebral oxygen desaturation is associated with early postoperative neuropsychological dysfunction in patients undergoing cardiac surgery. J Cardiothorac Vasc Anesth. 2004;18:552–8
6. Goldman S, Sutter F, Ferdinand F, Trace C. Optimizing intraoperative cerebral oxygen delivery using noninvasive cerebral oximetry decreases the incidence of stroke for cardiac surgical patients. Heart Surg Forum. 2004;7:E376–81
7. Fischer GW, Lin HM, Krol M, Galati MF, Di Luozzo G, Griepp RB, Reich DL. Noninvasive cerebral oxygenation may predict outcome in patients undergoing aortic arch surgery. J Thorac Cardiovasc Surg. 2011;141:815–21
8. Murkin JM, Adams SJ, Novick RJ, Quantz M, Bainbridge D, Iglesias I, Cleland A, Schaefer B, Irwin B, Fox S. Monitoring brain oxygen saturation during coronary bypass surgery: a randomized, prospective study. Anesth Analg. 2007;104:51–8
9. Slater JP, Guarino T, Stack J, Vinod K, Bustami RT, Brown JM 3rd, Rodriguez AL, Magovern CJ, Zaubler T, Freundlich K, Parr GV. Cerebral oxygen desaturation predicts cognitive decline and longer hospital stay after cardiac surgery. Ann Thorac Surg. 2009;87:36–44
10. Heringlake M, Garbers C, Käbler JH, Anderson I, Heinze H, Schön J, Berger KU, Dibbelt L, Sievers HH, Hanke T. Preoperative cerebral oxygen saturation and clinical outcomes in cardiac surgery. Anesthesiology. 2011;114:58–69
11. Pennekamp CW, Bots ML, Kappelle LJ, Moll FL, de Borst GJ. The value of near-infrared spectroscopy measured cerebral oximetry during carotid endarterectomy in perioperative stroke prevention. A review. Eur J Vasc Endovasc Surg. 2009;38:539–45
12. Lewis SB, Myburgh JA, Thornton EL, Reilly PL. Cerebral oxygenation monitoring by near-infrared spectroscopy is not clinically useful in patients with severe closed-head injury: a comparison with jugular venous bulb oximetry. Crit Care Med. 1996;24:1334–8
13. Mittnacht AJ. Near infrared spectroscopy in children at high risk of low perfusion. Curr Opin Anaesthesiol. 2010;23:342–7
14. Davie SN, Grocott HP. Impact of extracranial contamination on regional cerebral oxygen saturation: a comparison of three cerebral oximetry technologies. Anesthesiology. 2012;116:834–40
15. Sørensen H, Secher NH, Siebenmann C, Nielsen HB, Kohl-Bareis M, Lundby C, Rasmussen P. Cutaneous vasoconstriction affects near-infrared spectroscopy determined cerebral oxygen saturation during administration of norepinephrine. Anesthesiology. 2012;117:263–70
16. Kim MB, Ward DS, Cartwright CR, Kolano J, Chlebowski S, Henson LC. Estimation of jugular venous O2 saturation from cerebral oximetry or arterial O2 saturation during isocapnic hypoxia. J Clin Monit Comput. 2000;16:191–9
17. Feiner JR, Severinghaus JW, Bickler PE. Dark skin decreases the accuracy of pulse oximeters at low oxygen saturation: the effects of oximeter probe type and gender. Anesth Analg. 2007;105:S18–23
18. Matta BF, Lam AM. The rate of blood withdrawal affects the accuracy of jugular venous bulb. Oxygen saturation measurements. Anesthesiology. 1997;86:806–8
19. Bland JM, Altman DG. Agreement between methods of measurement with multiple observations per individual. J Biopharm Stat. 2007;17:571–82
20. Bickler PE, Feiner JR, Severinghaus JW. Effects of skin pigmentation on pulse oximeter accuracy at low saturation. Anesthesiology. 2005;102:715–9
21. Henson LC, Calalang C, Temp JA, Ward DS. Accuracy of a cerebral oximeter in healthy volunteers under conditions of isocapnic hypoxia. Anesthesiology. 1998;88:58–65
22. Ito H, Ibaraki M, Kanno I, Fukuda H, Miura S. Changes in the arterial fraction of human cerebral blood volume during hypercapnia and hypocapnia measured by positron emission tomography. J Cereb Blood Flow Metab. 2005;25:852–7
23. Ito H, Kanno I, Fukuda H. Human cerebral circulation: positron emission tomography studies. Ann Nucl Med. 2005;19:65–74
24. Willie CK, Macleod DB, Shaw AD, Smith KJ, Tzeng YC, Eves ND, Ikeda K, Graham J, Lewis NC, Day TA, Ainslie PN. Regional brain blood flow in man during acute changes in arterial blood gases. J Physiol. 2012;590:3261–75
© 2013 International Anesthesia Research Society
25. Butwick A, Hilton G, Carvalho B. Non-invasive haemoglobin measurement in patients undergoing elective Caesarean section. Br J Anaesth. 2012;108:271–7