Accuracy of a Cerebral Oximeter in Healthy Volunteers under Conditions of Isocapnic Hypoxia 

Thirty healthy volunteers (11 women, 19 men; aged 19–40 yr) were paid for their participation in these experiments. The study protocol was approved by the institutional review board. Volunteers gave written informed consent before undergoing preliminary testing, which included a medical history, physical examination, resting electrocardiogram, and measurement of hematocrit concentration. Pregnancy in women was excluded by medical history. Subjects were studied after a minimum 4-h fast.

The electrocardiogram and finger pulse oximeter (Siemens Sirecust 404, Danvers, MA) were monitored continuously for volunteer safety; blood pressure was measured noninvasively before and after placement of intravascular catheters, at the end of each experiment, after removal of the catheters, and as indicated throughout the experiments. Each volunteer was placed in the supine position and, after local anesthesia with 1% lidocaine, a 20-gauge or 22-gauge catheter was placed in either the right or left radial artery. Each volunteer was then placed in the Trendelenburg position and a 20-gauge, 5-inch catheter (Arrow International, Reading, PA) was placed retrograde in the right internal jugular vein, using routine sterile techniques, local anesthesia with 1% lidocaine, and ultrasound guidance (Site Rite, Dymax, Pittsburgh, PA) to identify the vein. The catheter was advanced until the volunteer noted a sensation in the jaw, indicating that the tip of the catheter was in the jugular bulb. The straight, retrograde course of the catheter to the mastoid process was confirmed using ultrasound. The volunteer was positioned supine for the remaining experiments. The cerebral oximeter probe (model INVOS 3100-SD, Somanetics Corp.) was placed on the right forehead, with the caudad border approximately 1 cm above the eyebrow with the medial edge at the midline. This position places the light source and sensors away from the frontal sinus. The oximeter algorithm averages data over 4-s intervals and the device displays a running 20-s average (i.e., an average of five 4-s values).

Volunteers breathed from a high-flow gas mixing chamber through a low dead-space face mask (Vital Signs, Totawa, NJ). Inhaled and exhaled gas volumes (Sensor Medics VMM 11, Laguna Hills, CA) and gas concentrations (model MGA 1100 mass spectrometer; Perkin-Elmer, Pomona, CA) were recorded breath by breath by computer using the TIDAL software package. [18]The flow meter and mass spectrometer were calibrated before each volunteer's experiments with volume and gas reference standards. The calibration of the cerebral oximeter was set at the factory and was not altered for any of the experiments.

The technique of dynamic end-tidal forcing was used to control P^ETO²and P^ETCO²tensions. [19]With this technique, a computer-driven gas mixing system adjusts the inspired concentrations of carbon dioxide and oxygen on a breath-by-breath basis. As ventilation changes, the system alters the composition of the inspired gas to maintain end-tidal gas tensions at constant values. A combination of predictive and adaptive control algorithms in the computer software allows precise regulation of end-tidal gas tension over prolonged periods and can also accomplish abrupt transitions in P sub ET O²or P^ETCO².

Volunteers first completed an experiment in which they breathed room air through the face mask for 10 min; P^ETCO²was averaged by the computer and the mean P^ETCO²for the last min was designated as that volunteer's resting P^ETCO². No blood samples were drawn during this 10-min period. Volunteers then completed two experiments with at least 30 min of rest between them. The first 30 s of each experiment was used to initialize the control algorithm; volunteers breathed room air from the gas mixing system. Then P^ETO²was controlled in the same sequence for all experiments: 80 mmHg for 6 min, followed by 4-min intervals at 45 mmHg, 60 mmHg, 51 mmHg, and 41 mmHg. For the final 4 min FIO²was controlled to 0.50. The O²tensions for these six test stages were designed to produce arterial saturation ranging from [approximately] 70% to 100%. After the first 30 s of room air breathing, P^ETCO²was controlled to [approximately] 2 mmHg greater than resting (experiment 1, normocapnia) or to 7–10 mmHg greater than the P^ETCO sub 2 used during the first experiment (experiment 2, hypercapnia).

During the last 30 s of each test stage of controlled O²tension, arterial and jugular venous blood samples were drawn over 10 s into syringes prepared with heparin and the cerebral oximeter reading (rSO²) was recorded simultaneously. Blood samples were immediately stored on ice and within 2 h were analyzed in duplicate by co-oximetry (model IL282; Instrumentation Laboratories, Lexington, MA). The co-oximeter was calibrated at the beginning of each day and as needed during the experiments.

There is no “gold standard” to use in assessing the accuracy of this device; i.e., there is no actual measurement of brain tissue oxygen saturation within the field of the sensor to compare with readings made by the cerebral oximeter. There are, of course, multiple sources for differences between the jugular venous saturation and the actual brain tissue oxygen saturation, including regional inhomogeneity of cerebral oxygenation, inaccuracy in the co-oximeter, and contamination of the jugular bulb sample with extracranial blood. There are also sources of error between the oximeter's reading and the actual tissue saturation in the field of the sensor, including contamination of the signal with extracranial absorption, intracranial arterial blood in the field, and inaccuracies in the signal processing and algorithms in the oximeter. Previous studies [16,17,20]have compared the oxygen saturation measured by the device (rSO²) to a calculated “field saturation”(S^fO²). The S^fO²is derived from the arterial and the jugular venous saturations (Equation 1), based on the assumption that the distribution of blood in the cerebral vasculature is three quarters in the venous bed and one quarter in the arterial bed, the same as in the rest of the body. [21–23]This assumption introduces a source of error in addition to those listed above.

By substituting rSO²(from the cerebral oximeter) for S sub f O²and using the measured SaO², this equation can also be used to estimate the jugular venous O²saturation (S^jvwith bar O²)(Equation 2). We compared the cerebral oximeter reading, rSO², and the derived S^jvwith bar O²(Equation 2), with the directly measured blood S^jvwith bar O². We also compared S^jvwith bar O²with SaO²because, under conditions of constant cerebral oxygen consumption and cerebral blood flow, S^jvwith bar O²tracks changes in SaO². Use of the measured jugular bulb blood saturation as the standard assigns most of the error to the cerebral oximeter reading and thus should provide a conservative evaluation of the accuracy of the device.

The cerebral oximeter with two wavelengths estimates only oxy- and deoxyhemoglobin, whereas the co-oximeter with four wavelengths also measures methemoglobin and carboxyhemoglobin. Therefore, to compare the readings made by the cerebral oximeter with co-oximeter measurements of blood samples, percentage O²hemoglobin (%O²Hb) readings from the co-oximeter were corrected for methemoglobin and carboxyhemoglobin (%O²Hb/[100 -%COHb -%MetHb]). This correction was used for arterial and jugular bulb samples; i.e., for SaO²and S^jvwith bar O². The average of the duplicate analyses of each blood sample was used in all calculations.

The relations of SaO², rSO², or S^jvO²to S^jvwith bar O²were determined by simple linear regression for each volunteer and for all volunteers combined (i.e., the pooled data for both experiments). The comparison of rSO²or S^jvwith bar O²to S^jvwith bar O²for the pooled data was also made by examining the difference plotted against the average of the two values, as described by Bland and Altman. [24]The error was expressed as bias and precision, where bias was calculated as the mean difference between rSO²or S^jvwith bar O²and S^jvwith bar O²and precision was calculated as 2 SDs of this difference. The significance of differences between mean values for different test stages or experiments was determined using paired t tests; for repeated t tests, the Bonferroni method was used to correct the accepted level of significance. A P value of 0.05 was considered significant.

All 30 volunteers completed the experiments without complications. No measurements were made from the cerebral oximeter for 2 of 360 test stages due to an inability of the device to capture a signal; both of these missing readings occurred in the same volunteer during the hypercapnic experiment. All arterial and jugular venous samples were obtained and analyzed. Samples representing the two cerebral oximeter readings that were not obtained were not used in subsequent analyses.

(Figure 1) shows a breath-by-breath record of the control of P^ETO²and P^ETCO²and the resulting pulse oximeter and cerebral oximeter readings for experiment 2 (hypercapnia) for a representative volunteer. With the exception of brief periods at the transients from one level of P^ETO²to another, P^ETO sub 2 and P^ETCO²were tightly controlled and were stable when readings from the cerebral oximeter and blood samples were obtained (the last 30 s of each test stage). Note that the finger pulse oximeter reading (SpO²) exhibited the expected delay of [approximately] 60 s after each transition in end-tidal (and therefore arterial) saturation. [25]The rSO²value from the cerebral oximeter also changed as predicted in response to changes in SaO². Although the transitions in rSO²were slower than the SpO²transitions, the cerebral oximeter had reached a stable reading by the end of each 4-min step in P sub ET O².

The P^ETCO²averaged 36.4 +/- 3.5 mmHg (mean +/- SD) for experiment 1 (normocapnia) and 43.1 +/- 4.7 mmHg for experiment 2 (hypercapnia). There were no significant differences among P^ETCO sub 2 values for the different test stages within either experiment. There were no significant differences in blood SaO²between the normocapnic and hypercapnic experiments for any test stage. The purpose of our protocol was achieved, in that blood S^jvwith bar O²was significantly higher in hypercapnia than in normocapnia at every SaO sub 2 (test stage). The rSO²tracked the changes in S^jvwith bar O²and was significantly higher in hypercapnia than in normocapnia. Although rSO²and S^jvwith bar O²did not differ during hypercapnia, rSO²was significantly higher than measured S^jvwith bar O²during normocapnia for all test stages except hyperoxia. Table 1summarizes these data.

(Table 2) gives statistics for the linear regressions between measured S^jvwith bar O²and SaO², rSO², or S^jvwith bar O²(from Equation 2) for individual volunteers. When data for both levels of P^ETCO²were combined, the mean r sup 2 for the SaO²relation was 0.71. Mean r²was much higher (0.91) for rSO²and S^jvwith bar O²(i.e., for the data based on the cerebral spectroscopy measurement). However, the slopes and intercepts for all three relations varied widely among volunteers. For some, there were large differences between estimated and measured saturation, whereas for the other volunteers the values corresponded closely. As a result, for the pooled data on all volunteers, S^jvwith bar O²was only loosely correlated with any of the other measures (S^jvwith bar O²= 0.81 SaO²- 10.5%, r²= 0.55; S^jvwith bar O²= 0.63 rSO²+ 20%, r²= 0.44; S^jvwith bar O²= 0.45 S^jvwith bar O²+ 34.7%, r sup 2 = 0.30). Figure 2illustrates the error that can be introduced by this range of slopes and intercepts. This figure shows the data from three individual volunteers; as can be seen, an rSO²reading of 65% corresponds to a jugular venous saturation of 48%, 58%, or 66%, depending on the volunteer.

A Bland-Altman plot of the error of the cerebral oximeter reading, rSO², revealed two “populations” of data (Figure 3). Most of the data clustered within +/- 2 SDs of the bias (the mean error), and for this subset of the data the error did not vary systematically with the mean of the two measurements. However, there was a smaller subset of data with very large negative error (i.e., the cerebral oximeter grossly underestimated the measured S^jvwith bar O²). Most of these data were obtained in three volunteers for whom the cerebral oximeter underestimated S^jvwith bar O²at all levels of SaO²and both levels of CO²(indicated by the triangles in Figure 3). For individual volunteers, bias ranged from -40% to +10% and precision ranged from 3% to 64%. For all, the data bias was 3.8%, and the precision was 17.8%. When the comparison was made with S^jvwith bar O², the bias was -4.4% with a precision of 22%. The error was not correlated with either P^ETCO²(r²= 0.05) or arterial saturation (r²= 0.02).

The cerebral oximeter assessed in this study (model INVOS 3100A, Somanetics Corp.) was recently approved by the US Food and Drug Administration for clinical use. Ours is the first validation study of this device in healthy volunteers to tightly control end-tidal (and therefore arterial) tensions of O²and CO². The dynamic end-tidal forcing technique allowed us to step rapidly from one level of P^ETO²to another and maintain a constant level of saturation for 4–6 min while also controlling P^ETCO². By regulating end-tidal gas tensions, we have tried to ensure that the degree of cerebral vasodilatation secondary to Pa^CO2and the arterial oxygen saturation were constant at the time of measurement. Previous studies have used fixed inspired concentrations of O²without control of CO²[17]; fixed FIO²with voluntary hyperventilation, CO²rebreathing, or both [16]; or have used a constant low FIO²resulting in “ramp” changes in SaO². [20]Consequently, cerebral oximeter readings and blood sampling may have occurred in a changing cerebral oxygenation “environment.”

Validation of the cerebral oximeter is difficult because there is no way to directly measure regional brain oxygenation in the field of oximeter probe. Previous studies have compared readings from the cerebral oximeter to a “field” or “combined” saturation calculated from a combination of arterial and ipsilateral jugular venous bulb saturation. [16,17,20]This calculated value (0.25 [center dot] SaO sub 2 + 0.75 [center dot] S^jvwith bar O²) is based on the assumption that the distribution of blood in the cerebral vasculature is the same as that in the rest of the body [22]and the fact that approximately three-quarters of the peripheral blood volume is usually stored in the peripheral venous bed. [21]Because few data support this stated proportional distribution of arterial and venous blood, we compared the measured jugular venous saturation to both the adjusted and unadjusted oximeter readings. The correction did not significantly improve either the accuracy or the correlation of the measurement. In interpreting the data, it should be remembered that the jugular bulb receives some venous blood from the contralateral brain but almost none from the posterior fossa [26]and may reflect the middle cerebral artery more than the anterior cerebral artery. Therefore, whereas jugular venous bulb saturation may reflect global cerebral oxygen delivery in relation to demand, it may not accurately reflect regional saturation in the cortical tissue of the frontal lobe of the brain (i.e., the region underlying the oximeter probe).

If we assume that global cerebral oxygen consumption does not change during the period of measurement, jugular bulb saturation should decrease with decreased arterial saturation or decreased Pa^CO2due to decreased oxygen delivery. In 1942 Nims et al. [27]reported that a 1 vol% change in CO²content of the arterial blood produced a 0.5 vol% change in O²content of jugular venous blood in three healthy persons; the relation was approximately linear over a wide range of hypo- and hypercapnia. As expected, we found that measured S^jvwith bar O²was lower with lower SaO²for the same P^ETCO sub 2 and higher with hypercapnia than with normocapnia for the same SaO sub 2.

Because the saturation of hemoglobin in extracranial tissue in the light path of the oximeter probe is expected to be one of the major contaminants producing errors in the measurement, it is important to carefully examine the errors at a constant SaO²for different P sub ET CO². Changes in P^ETCO²at constant arterial saturation would be expected to change S^jvwith bar O²without changing the hemoglobin saturation in extracranial tissue. We obtained the same target SaO²for all test stages (Table 1). The oximeter reading increased with hypercapnia, indicating that the cerebral oximeter is responding to changes in intracranial hemoglobin oxygen saturation. However, the cerebral oximeter reading was significantly higher than blood S^jvwith bar O²for normocapnia for all P sub O²values except hyperoxia, whereas the cerebral oximeter reading did not differ from S^jvwith bar O²during hypercapnia. These observations could indicate either that not all the contamination from extracranial tissue has been eliminated or that the ratio of arterial to venous blood in the light path is altered by hypercapnia. Because hypercapnia is a potent cerebral vasodilator, the latter is a likely possibility. Interestingly, Pollard et al. [16]found that the decrease in the cerebral oximeter reading was smaller than the decrease in the calculated “combined” saturation under conditions of hypocapnia in a small number of healthy volunteers.

Several case reports and small clinical series have indicated that cerebral oximetry tracks expected changes in cerebral oxygen saturation with changes in arterial blood pressure or oxygen saturation, [9]hypothermic circulatory arrest, [10]carotid artery compression, [28]and subsequent morbid clinical events. [11,29]It has therefore been suggested that cerebral oximetry may be useful for clinical monitoring of brain oxygen supply during various procedures. [7,8,12,30]However, few published studies have addressed the accuracy of the newly approved device in humans and only two previous reports have considered its accuracy under conditions of controlled cerebral metabolism and oxygen delivery in healthy human volunteers. [16,17] 

Dujovny et al., [31]using a prototype of the current device that had a shorter emitter-detector spacing, reported no relation between the baseline reading and age in 100 persons studied. McCormick et al. [20]compared readings from that same device with measured arterial and jugular venous bulb saturations in nine patients in the neurologic intensive care unit. Oximeter readings correlated poorly (r sup 2 = 0.55, n = 68) with “field” saturation calculated from two models of cerebral blood distribution (arterial:venous = 0.18:0.82 or 0.28:0.72). McCormick et al. [20]also tested the device in seven healthy persons with ramp changes in SaO²induced by breathing 7% O sub 2 but did not collect jugular venous blood samples and therefore could not perform the correlation analysis for the healthy persons. The short emitter-detector distance was subsequently lengthened to the current 30 mm and 40 mm in an attempt to reduce extracranial contamination of the saturation reading.

Pollard et al. [17]studied the response of the newly approved device (model INVOS 3100A) to arterial hypoxemia in 22 healthy persons. Hypoxemia was induced by having volunteers breathe hypoxic gas mixtures until a stable pulse oximeter reading was obtained. The P^ETCO²was not controlled and was presumably lower with greater degrees of hypoxia. The cerebral oximeter reading correlated well with calculated “combined” saturation; r²(the coefficient of determination) ranged from 0.79 to 0.99 in individual volunteers. Bias ranged from -10% to +8% in individuals; for the group of volunteers, bias was -0.7%. Precision (defined as +/- 2 SDs about the mean error) ranged from 2% to 8% in individual volunteers; for the group data, precision was 10.4%. We observed both similarly high r²values for individuals and similar variance in individual bias and precision in our study; however, precision of the group data was poorer (2 SDs = 22%). Our calculated error for jugular venous saturation will be 50% larger than the calculated error for field saturation.

Pollard et al. [16]also examined the effects of body position (20% Trendelenburg, supine, or 20% reverse Trendelenburg) with and without varying P^ETCO²on the accuracy of the INVOS 3100A cerebral oximeter during arterial hypoxemia. Carbon dioxide was designated “normal”(i.e., P^ETCO²not controlled, as in the previous study), “high”(volunteers inhaled CO²to increase P^ETCO²to > 55 mmHg), or “low”(volunteers voluntarily hyperventilated to P^ETCO²< 15 mmHg). Jugular venous saturation, “combined” saturation, and cerebral oximeter saturation were significantly higher for the Trendelenburg position and hypercapnia (which should increase cerebral blood volume, flow, or both) and were significantly lower for hypocapnia. However, the decrease in the cerebral oximeter reading was smaller than the decrease in the calculated “combined” saturation under conditions of hypocapnia. In this study, mean precision (2 SDs) for eight volunteers studied with varying CO²was 17%, which is higher than in their poikilocapnic study [17]and similar to the precision in our study for 30 persons with controlled normocapnia and hypercapnia.

Brown et al. [32]compared the cerebral oximeter and continuous measurements of jugular bulb saturation (Oximetrix Opticath, Abbott Laboratories, North Chicago, IL) to measured S^jvwith bar O sub 2 in nine patients undergoing procedures involving cardiopulmonary bypass; data were collected throughout surgery, including before and after bypass. The cerebral oximeter was much less precise (2 SDs = 28.2%) than the Opticath (2 SDs = 5.4%), and error of the cerebral oximeter varied systematically over the range of measured S^jvwith bar O²([approximately] 40% to [approximately] 95%). This difference was attributed to changes in partitioning of arterial, capillary, and venous blood in the brain during nonpulsatile flow, which would presumably affect the cerebral oximeter but not the saturation of jugular venous blood. In individual healthy volunteers, the cerebral oximeter reading correlates well with measured “combined” or “field” saturation and does not exhibit systematic bias [16,17](this study). Interestingly, Trubiano et al. [33]recently reported poor precision (2 SDs =[approximately] 42%) of the Opticath in a similar clinical situation. It is unclear why the accuracy of the Opticath was so vastly different between these two studies.

Our major finding is that S^jvwith bar O²can be estimated by the INVOS 3100A using the uncorrected direct measurement rSO²or by correcting the measurement using the measured SaO²(Equation 2) under conditions of hypoxic exposure with controlled normocapnia and hypercapnia. This implies that the device detects changes in cerebral oxygen saturation, due to changes in cerebral blood flow induced by changes in P^ETCO², which are not apparent from the arterial saturation. However, the wide variability among volunteers of the slope and intercept of the relation between the actual jugular venous saturation and the estimate from the device, as well as the occurrence of three clear outliers (Figure 2), limits the clinical use of the device to situations in which tracking trends in cerebral oxygenation would be acceptable. An improved algorithm in the device has eliminated the problem with outliers (Henson LC, Ward DS. Unpublished data). Although cerebral oximetry appears to hold promise for the noninvasive estimation of brain oxygenation, the utility and accuracy of the technique in clinical situations in which cerebral oxygen saturation is reduced because of pathologic causes (such as reduced flow) remains to be shown.