Effects of Changes in Arterial Carbon Dioxide and Oxygen Partial Pressures on Cerebral Oximeter Performance

CEREBRAL oximetry is a noninvasive optical monitor of brain oxygenation that integrates arterial and venous oxyhemoglobin saturation in cortical tissue using near-infrared spectroscopy.^1,2  The volumetric contribution of cerebral capillary blood is minimal (less than 2%).³  Clinical studies have examined the use of cerebral oximetry for goal-directed care of patients requiring cardiopulmonary bypass,⁴  with decreased oximeter readings correlating with an increased incidence of neuropsychiatric impairment,⁵  stroke,⁶  organ dysfunction, and mortality.^7,8  Other studies have examined cerebral oximetry for detection of cerebral hypoperfusion during carotid endarterectomy,⁹  adequacy of advanced cardiac life support,¹⁰  and directing care of patients with traumatic brain injury.¹¹  However, case reports describe ischemic events occurring without changes in cerebral oximeter signals¹²  and potential for false-positive cerebral oximeter desaturation events not correlated with adverse neurologic sequelae in beach chair position.^13,14  In addition, there is currently poor correlation between cerebral oximeter desaturation events and the incidence of adverse neurologic outcomes in cardiac surgery patients and traumatic brain injury.^15,16 

Cerebral oximetry relies on assumptions requiring additional study to determine the degree of measurement limitation. Cerebral oximetry measures the ratio of oxyhemoglobin to total hemoglobin (oxy- plus deoxy-) within cortical tissue as an indirect estimate of regional brain tissue oxygenation (cerebral oxygen saturation; ScO₂). Near-infrared light is transmitted through the forehead with two sets of photodetectors used to capture light from different tissue depths. Although it is assumed that this allows discrimination of signal from shallow tissues (i.e., scalp, galea, or bone sinuses) in favor of deeper brain tissue measurement, the actual degree of extracranial contamination is unclear and likely varies significantly between devices and associated algorithms.¹⁷  Additional assumptions of cerebral oximetry include uniform patient tissue geometry, metabolic rate, and a constant ratio of venous to arterial blood volume (e.g., 70% venous/30% arterial) in the sensor field using values derived from previous cerebral positron emission tomography volumetric imaging studies.¹⁸ 

Although previous human studies evaluated the performance of cerebral oximetry in relation to hypoxia,^2,19,20  possible error in measurement from varying Paco₂ has not been examined. Changes in Paco₂ alter the relative volumes of arterial and venous blood in the brain²¹  and may create significant error with the assumption of a fixed venous/arterial volume ratio, because a 10-mmHg decrease in Paco₂ can decrease the intracranial arterial blood volume by 50%.³  Compared with normoxemia, hypoxemia (80% arterial saturation) with isocapnia results in a significant increase in middle cerebral artery flow velocity (approximately 25%) that is attenuated with hypocapnia.²²  Various degrees of hypocapnia combined with hypoxia may individually alter cerebral blood flow (CBF),²³  highlighting the need to examine changes in both simultaneously. Although studies have noted a direct relationship between ScO₂ and Paco₂,^24–27  we are unaware of studies in which oximeter accuracy and bias were examined in relation to an acute change in Paco₂ during concurrent changes in Pao₂. This has clinical significance because patients at risk for cerebral ischemia, those with brain injury, or those undergoing neurosurgical procedures are often purposefully held in a hypocapnic state with mechanical ventilation. Hyperoxia is relevant because patients routinely receive ventilation with high inspired oxygen levels in the operating room or intensive care unit (ICU), which often results in a high Pao₂ unless there is underlying pulmonary dysfunction.

Clinical trials with cerebral oximetry often include interventions that manipulate inspired oxygen and minute ventilation. This may significantly alter total and regional CBF patterns with unclear effects on the underlying theoretical measurement assumptions.^28,29  Accurate interpretation of cerebral oximetry values relies on knowing whether these interventions result in a true increase in brain oxygen levels or are an artifact from extracranial signal contamination or changes in the ratio of arterial and venous cerebral volume. Accordingly, the purpose of this healthy volunteer investigation was to assess the accuracy and limitations of cerebral oximetry under simulated clinical conditions with changes in both Paco₂ and Pao₂. We hypothesized that hypocapnia in combination with extremes of inspired oxygen would decrease both the accuracy and precision of cerebral oximetry.

All of the protocols were approved by the University of California at San Francisco Committee on Human Research (San Francisco, California) and written informed consent obtained before any intervention. Methods were similar to previous evaluation of cerebral oximeter accuracy by Bickler et al.²  Thirteen subjects were enrolled in this study. One subject completed less than half of the breathing protocol due to nausea and was eliminated from analysis. Of the 12 subjects completing the study, there were eight men and four women of various ethnicities (nine white, one Asian, and two black) and a range of skin tones (half lighter and half darker). Subjects were between the ages of 23 and 30 yr of age (mean ± SD, 26 ± 2 yr) with average body mass index (24.8 ± 3.9 kg/m²). Preenrollment screening with a history and physical examination excluded cardiopulmonary conditions that could affect measurements and safety. Pregnancy was excluded by history. Data were collected between October 2013 and February 2014.

A 22-gauge radial arterial catheter was placed to monitor blood pressure and sample arterial blood. A right internal jugular venous catheter was inserted under ultrasound visualization using a Seldinger technique with the wire directed cranially. A 5-cm, 20-gauge catheter was threaded over the wire and advanced toward the jugular bulb. The intravascular location of the catheter and position of the catheter tip within the internal jugular bulb was confirmed by ultrasound and adjusted if necessary.

The accuracy of two commercial cerebral oximeters (FORE-SIGHT Elite; CAS Medical Systems, Inc., USA; and INVOS 5100C; Covidien, USA) was evaluated under two Paco₂ levels and a range of fractional inspired oxygen (FIO₂). An oximeter from each manufacturer was placed on the forehead symmetrically to either side of midline with maximal separation as allowed by the subject’s anatomy. Right or left position was determined by previous subject randomization. Sensors were covered with a dark opaque cloth to minimize signal contamination by ambient light.

Arterial and venous blood were withdrawn simultaneously. A 1-ml jugular venous sample was obtained over a period of 30 s to prevent nonjugular bulb reflux into the catheter from the distal circulation.^2,30  Arterial and venous gases were analyzed using either the ABL 800 or ABL 90 multiwavelength blood gas analyzers (Radiometer Medical, Denmark), because the device available changed during the study period. Daily calibrations were performed to ensure accuracy. Consistency between the devices was ensured by concurrent analysis of a subset of arterial and venous blood samples before use in our protocol, and only one device was used on any particular subject.

Blood samples were analyzed to determine arterial and venous oxygen saturations, Pao₂ and Paco₂. A weighted reference saturation was calculated for each pair of samples using the fixed ratio of arterial to venous intracranial blood described previously by the device manufacturers. Weighting ratios are 25%/75% arterial/venous for the INVOS and 30%/70% for the FORE-SIGHT. Heart rate, arterial blood pressure, peripheral capillary oxygen saturation, FIO₂, and end-tidal carbon dioxide were also continuously monitored throughout the study and recorded at each point of blood gas analysis. All of the data were collected and stored using a LabVIEW (National Instruments, USA) interface, which allowed temporal resolution from the various continuous monitoring devices. Continuous data values used in the analysis (e.g., ScO₂) were averaged over a 10-s period spanning the timestamp that marked the initiation of each arterial blood draw.

Baseline data were obtained with subjects breathing room air. The volunteers were then exposed to set levels of FIO₂, including 1.0, 0.8, 0.6, 0.4, 0.21 (room air), 0.085, and 0.06, using a semiopen breathing system that allowed each FIO₂ to be held constant for the necessary period of equilibration. The two hypoxic levels (FIO₂ approximately 0.085 and 0.06) were titrated to target an arterial oxygen saturation (Sao₂) of 90% and 80%, respectively. To evaluate the effect of varying Paco₂, the same set of FIO₂ measurements was performed under conditions of both normocapnia and hypocapnia. We targeted a Paco₂ of 40 and 30 mmHg, respectively, with an analysis cutoff of 35 mmHg. Any sample outside the intended Paco₂ range was reclassified for analysis. Hypocapnia was induced by coached subject hyperventilation, and normocapnia required the addition of carbon dioxide in the inspired gas flow. Paco₂ levels were maintained constant at the desired target level by continuously monitoring end-tidal carbon dioxide, which was then compared with the Paco₂ blood gas samples to determine a relative difference and allow small corrections in subject breathing or exogenous inspired carbon dioxide as needed. To negate any sequential effects, subjects were randomly assigned to both the order of Paco₂ studied (normocapnia first or hypocapnia first) and the order in which the FIO₂ was varied (increasing FIO₂vs. decreasing FIO₂).

The study was powered based on previous experience with similar repeated-measures analyses for cerebral oximeters.²  A P value less than 0.05 was considered significant. An a priori power calculation based on previous data²  and pilot studies of paired differences in ScO₂ between step changes in Sao₂ or Paco₂ determined that we would need seven subjects to detect an average difference of 3% ScO₂, based on the SD of differences determined to be less than 2%. Stated another way, the use of repeated measures is very robust and powerful for these data because measured changes in ScO₂ within subjects is very consistent. Based on previous experience, we used 12 subjects to increase our sensitivity to detect smaller differences. The repeated-measures regression used for our primary analysis is more robust because it combines all data, using a continuous measurement scale. Statistical analyses were performed with JMP 11.0 (SAS Institute, USA) and Stata 14 (StataCorp, USA).

Oximeter bias from each paired arterial and venous blood sample was computed as the difference between the displayed oximeter reading and the weighted reference saturation described above. Bland–Altman plots were constructed displaying the bias as a function of the weighted reference saturation under both normocapnic and hypocapnic conditions. In addition, the following standard statistics associated with device performance were calculated across the range of FIO₂ examined: mean bias, SD, 95% limits of agreement (LOA) adjusted for repeated measures,³¹  and average root mean square error (A_RMS). The 95% CIs for A_RMS and LOA were calculated using bootstrapping (random resampling with replacement) with 50,000 repetitions. These measures were calculated for each instrument under both normocapnic and hypocapnic conditions, as well as a composite.

Comparison of bias for a given carbon dioxide level within FIO₂ targets or within reference saturation ranges was determined using repeated-measures ANOVA. Tukey–Kramer honestly significant difference was used for multiple comparisons of values between the different reference saturation ranges. Because multiple comparisons would be excessive for seven different FIO₂ levels and the primary analysis was a linear regression, these were not performed for the FIO₂ groups. A paired t test was used for comparison of values between high and low Paco₂.

Linear regression lines were constructed for the bias plots using a repeated-measures analysis. The repeated-measures analysis was performed as a linear mixed-effects analysis using the subject identification as a random effect. Regression of ScO₂versus the reciprocal of Pao₂ using the mixed-effects model fit the data well, with additional variables not significantly improving fit. This regression was analyzed separately for normocapnic or hypocapnic groups with respect to repeated measures. Separate analysis of values near full Sao₂ (FIO₂ ≥ 0.21) used a simple repeated-measures linear regression for ScO₂ and Svo₂. Linearity was examined by testing the statistical significance of a chi-squared term.

Arterial and venous samples were drawn simultaneously at each inspired oxygen level under both normocapnic and hypocapnic conditions for all 12 subjects. A total of 168 paired samples were analyzed.

Table 1 displays a summary of the oximeter and blood gas data for each of the target Paco₂ levels and range of specific FIO₂. Paco₂ was controlled within a narrow range, clearly separating the normocapnia and hypocapnia groups, with only 3 of the 168 paired samples requiring reclassification for not meeting the intended Paco₂. Mean Pao₂ for each of the target FIO₂ levels was similar when comparing analogous samples from the normocapnia and hypocapnia groups.

Although arterial blood pressure showed a statistically significant decrease with hypoxia, this change represented less than 4-mmHg difference in mean arterial pressure over the range of Pao₂ and Paco₂. However, we did observe a clinically relevant increase in heart rate at lower Pao₂ levels (not shown) consistent with previous studies addressing the hemodynamic effects of hypoxia.²² 

To examine the effect of Paco₂ and Pao₂ on ScO₂, displayed values from both the FORE-SIGHT and INVOS measurement systems were plotted as a function of Pao₂ over the range of FIO₂ for both normocapnia and hypocapnia (fig. 1, A and B, respectively). Both the INVOS and FORE-SIGHT ScO₂ plots were fit by the same nonlinear inverse function (see figures for equations). ScO₂ increased in a curvilinear manner with increasing Pao₂, with modest but significant increases in the ScO₂ above a Pao₂ of approximately 100 mmHg.

For FIO₂ of room air and higher, data were analyzed by linear regression, because the chi-squared term was not statistically significant. From room air to the hyperoxic Pao₂ range, the ScO₂ continued rising at higher FIO₂ for both FORE-SIGHT and INVOS (P < 0.0001). Hypocapnia resulted in decreased ScO₂ values as compared with normocapnia across the range of inspired oxygen concentrations for both the FORE-SIGHT (P < 0.0001) and INVOS (P = 0.0061) monitors. However, the slopes were not statistically different between normal and low Paco₂.

Figure 1, C and D, plot the weighted reference saturation against the Pao₂. The reference saturation was derived from the same blood gas values for both plots but using the 30%/70% arterial–venous weighting ratio for the FORE-SIGHT (fig. 1C) and the 25%/75% ratio for the INVOS (fig. 1D). Similar to the raw ScO₂ plots for each oximeter, both the FORE-SIGHT and INVOS weighted reference saturations increased in a curvilinear manner with increasing Pao₂ and were well fit with a simple inverse function. Hypocapnia resulted in a decrease in the weighted reference saturation over the range of Pao₂ for both the FORE-SIGHT and INVOS (P < 0.0001).

Accuracy was assessed using bias calculated by comparing the ScO₂ with a weighted saturation determined from the blood gas samples, as described previously. Table 2 displays oximeter accuracy data broken down by corresponding weighted saturation ranges for both normocapnic and hypocapnic conditions. Mean bias, precision, LOA, and A_RMS with corresponding CIs are shown for designated ranges of weighted reference saturations. Over the entire range of experimental conditions, the mean bias ± precision and A_RMS error were 2 ± 8% and 8% (95% CI, 7 to 9%) for the FORE-SIGHT and 4 ± 14% and 14% (95% CI, 13 to 16%) for the INVOS.

Figure 2 displays Bland–Altman plots of oximeter bias versus weighted reference saturation for both devices. As in previous studies, bias for the FORE-SIGHT monitor under conditions of normocapnia (fig. 2A) was greater at lower weighted reference saturations, indicated by the negative bias plot slope (slope –0.28; P < 0.0001). Stated differently, the FORE-SIGHT tended to overestimate the ScO₂ at lower saturations while underestimating it at higher saturations. Bias was also more positive during hypocapnia as compared with normocapnia (4 ± 9% vs. 0 ± 6%; P < 0.001). Hypocapnia altered the bias plot slope, –0.35 versus –0.28 for normocapnia (P = 0.010). The LOA and A_RMS were greater for hypocapnia than normocapnia, suggesting greater variability at low carbon dioxide.

The plot of bias as a function of weighted reference saturation for the INVOS monitor (fig. 2B) showed marginally negative slopes for both normocapnia (–0.03) and hypocapnia (–0.06) that were not significant. Bias was more positive during hypocapnia than normocapnia (5 ± 15% vs. 3 ± 12%; P = 0.010). Hypocapnia did not appreciably alter the relationship of bias to the weighted reference saturation with similar slopes between the two Paco₂ conditions. The LOA and A_RMS errors were larger for hypocapnia as compared with normocapnia. Both values were larger for the INVOS than for the FORE-SIGHT.

Figure 3 plots the effect of Pao₂ on bias to determine whether the oximeter accuracy varied as a function of increasing Pao₂. For the FORE-SIGHT (fig. 3A), bias decreased with increasing Pao₂ under normocapnic conditions (P < 0.001) but was relatively constant under hypocapnic conditions. As in the Bland–Altman plots, bias was more positive for hypocapnia as compared with normocapnia across the range of Pao₂ (P < 0.0001). The slope of the relationship of bias to Pao₂ was significantly more negative for normocapnia as compared with hypocapnia (P = 0.03). For the INVOS monitor (fig. 3B), bias remained relatively constant with increasing Pao₂ for normocapnia, whereas bias increased with increasing Pao₂ during hypocapnia (P < 0.0001). Again, bias was greater and more positive for hypocapnia than normocapnia (P = 0.004). The slope of the relationship between bias and Pao₂ was increased for hypocapnia as compared with normocapnia (P = 0.04).

Given the known effects of Paco₂ on CBF,²¹  we plotted jugular bulb oxygen saturation (Svo₂) as a function of Pao₂ under both normocapnic and hypocapnic conditions (fig. 4A). Svo₂ increased linearly with Pao₂ under conditions of both normocapnia (slope = 0.04; P < 0.0001) and hypocapnia (slope = 0.02; P < 0.0001). Hypocapnia caused a decrease in Svo₂ and flattening of the slope (P < 0.0001), suggesting greater oxygen extraction. Jugular venous oxygen content similarly increased linearly with Pao₂ under both normocapnic and hypocapnic conditions (not shown). Again, hypocapnia caused a decrease in jugular venous oxygen content over the range of Pao₂ and flattened the slope of the plot (not shown).

Overall oxygen extraction remained relatively constant with increasing Pao₂ (plot not shown). However, extraction was greater during hypocapnia (9.7 ± 2.7 ml/dl) as compared with normocapnia (6.9 ± 1.7 ml/dl; P < 0.0001). Oxygen extraction was normalized to arterial oxygen content to produce a fractional oxygen extraction (FOE) and eliminate the effect of baseline differences in oxygen content (fig. 4B). When FOE was plotted against Pao₂, the FOE decreased with increasing Pao₂ for normocapnia (P < 0.0001). Hypocapnia not only caused an increase in FOE over the entire range of Pao₂ but also negated the decrease in FOE noted at higher Pao₂ under normocapnic conditions. This increase in the slope for hypocapnia compared with normocapnia was significant (P < 0.0001).

Because many studies used Svo₂ as a surrogate for CBF and extraction, we plotted FOE as a function of Svo₂ (fig. 4C). FOE was negatively correlated with Svo₂ and increased as SvO₂ decreased (P < 0.0001). Hypocapnia magnified this relationship with a more negative slope (P = 0.005).

We next plotted oximeter bias as a function of FOE to assess for a relationship. For normocapnia, FORE-SIGHT bias (fig. 5A) increased with FOE (P < 0.0001). Hypocapnia resulted in a uniform decrease in the bias across the range of FOE (P < 0.0001) without changing the slope of the relationship.

For the INVOS oximeter (fig. 5B) under conditions of normocapnia, bias increased with increasing FOE (P < 0.0001). With hypocapnia, bias decreased as compared with normocapnia. The difference in the slopes between normocapnia and hypercapnia was not significant.

Our study shows that cerebral oximetry reflects changes in brain tissue oxygenation with increasing Pao₂, but coexisting hypocapnia causes a clinically significant decrease in ScO₂ with a moderate positive bias in the readings. This suggests that assumptions of a constant arterial–venous blood volume may be incorrect under certain conditions and that Paco₂ levels should be taken into account when interpreting ScO₂.

The relationship between Pao₂ level and cerebral oxygenation was curvilinear, similar to previous studies³²  (fig. 1). FIO₂ greater than 0.3 is common during surgery and ICU care. In the corresponding hyperoxic range, increases in cerebral saturation occurred with increasing Pao₂ despite arterial hemoglobin saturations remaining constant near 100%. This increase was likely driven by increased dissolved oxygen and greater cerebral venous saturation. Previous studies note a positive correlation between ScO₂ and cerebral venous saturation.¹⁹ 

The accuracy of ScO₂ estimates of cerebral oxygenation varied with the Pao₂. For the FORE-SIGHT, bias was positive at low saturations but progressively decreased as the saturation increased, becoming negative in the hyperoxic range (fig. 2A), thus overestimating the state of cerebral oxygenation with hypoxia while underestimating it with hyperoxia. This relationship did not exist for the INVOS monitor, possibly related to greater variability in the readings.

The relationship of bias to Pao₂ under conditions of normocapnia depended on the oximeter. For the FORE-SIGHT, bias decreased with increasing Pao₂, whereas for the INVOS there was no change as Pao₂ increased. However, the data variability for this INVOS relationship was large. In contrast, bias was closely correlated with FOE, likely from the higher weighting of the venous compartment in the oximeter output. This suggests that ScO₂ accuracy is perhaps more affected by hypocapnic changes in CBF rather than the Pao₂.

A shift from normocapnia to hypocapnia caused a decrease in ScO₂ across the range of measured Pao₂ for both oximeters (fig. 1). This is consistent with previous research demonstrating that ScO₂ increased in a linear fashion with increasing carbon dioxide during conditions of constant Pao₂,^24–27,32  although ScO₂ accuracy and bias were not systematically examined.

With hypocapnia, one would expect lower CBF and a decreased cerebral blood volume due to constriction of the cerebral vasculature.²⁸  In our study, hypocapnia resulted in increased FOE and a decreased cerebral venous oxygen content. These values correspond with physiologic changes expected from a decrease in CBF during hypocapnia. Both the ScO₂ and weighted reference saturation decreased during hypocapnia (fig. 1). However, ScO₂ decreased less than the reference saturation, resulting in a more positive bias for hypocapnia (compared with normocapnia) for both oximeters. When analyzing bias as a function of Pao₂ (fig. 3), hypocapnia increased both mean bias and the bias across all of the measured Pao₂ levels. However, when examining bias as a function of weighted reference saturation (fig. 2), part of this difference in mean bias appears to be driven by a clustering of the hypocapnic values at the lower reference saturations where bias is greater. This is consistent with previous accuracy studies under normocapnic hypoxia.²  Presumably the leftward clustering of hypocapnic values is due to increased FOE and the anchoring effect of the greater weighting of the venous fraction, resulting in a lower weighted reference saturation for any given Pao₂ with hypocapnia. This results in a significantly greater bias with hypocapnia.

Commercially available cerebral oximeters assume a fixed ratio of arterial to venous intracranial blood volumes. However, the specific fixed arterial–venous weighting ratio varies modestly between oximeter models as a result of differences in the signal processing algorithm and original device calibration studies (e.g., a 25%/75% fixed ratio is used for the INVOS, whereas a 30%/70% is used for the FORE-SIGHT). However, based on imaging studies, changes in CBF seen with altered Paco₂ levels manifest as changes in intracranial arterial blood volume without a significant change in venous blood volume.²⁹  This change in arterial blood volume is linear in relation to Paco₂, and a Paco₂ decrease of similar magnitude (40 to 30 mmHg) is demonstrated to create a twofold decrease in both the intracranial arterial blood volume and the fraction of arterial blood (46 to 23%) based on a study by Ito et al.³  Because the arterial–venous weighting ratio of the oximeter is fixed, we would expect decreases in arterial flow during hypocapnia to create significant error. Our data demonstrate that, with acute hypocapnia (which decreases brain oxygen levels), the oximeters tend to overestimate the state of cerebral oxygenation, a bias that is even greater at lower states of oxygenation.

A study of patients with traumatic brain injury with positron emission tomography imaging noted that hyperventilation reduced CBF, increased FOE, and increased ischemic brain volume.³³  Our study found substantially increased variability in the ScO₂ between subjects under conditions of hypocapnia. It may be that hypocapnia has significantly different effects on vascular tone between individuals. In addition, a complex interplay of CBF, oxygen extraction, and contamination from other signal sources may also be at work.²³  Hypocapnia resulted in a rightward shift and flattened slope of the plot of cerebral venous saturation versus Pao₂ (fig. 4A), suggesting greater FOE during the lower flow state. Our results (fig. 4B) also noted that, under normocapnic conditions, FOE decreased with increasing Pao₂, yet with hypocapnia this relationship no longer remained. This suggests that, under the lower flow hypocapnic state, oxygen extraction may have reached its limit. Approaching an extraction limit may also increase the ScO₂ measurement variability between individuals.

Finally, the contribution of extracranial tissue perfusion to the signal is unknown, particularly under conditions of hypocapnia and hypoxia. Although previously assumed to be small, recent studies have demonstrated a 7 to 17% decrease in ScO₂ readings when scalp cutaneous blood flow was manipulated by inflating a circumferential cranial pneumatic cuff.³⁴  The magnitude of this decrease varied depending on the oximeter used and was only partially attenuated by using a four-wavelength measurement present in newer-generation devices.¹⁷  Although cutaneous blood flow appears relatively unaffected by Paco₂,³⁵  blood flow may also be affected by the increases in heart rate and cardiac output seen with hypoxemia. Any extracranial flow differences may contribute a larger error, because hypocapnia decreases the overall CBF and individual differences may account for some of the between-subject variability.

We studied healthy volunteers rather than patients at risk for cerebral ischemia. Patients often have structural central nervous system abnormalities and perturbations in autoregulation of CBF that may alter cerebral oximetry in a manner distinct from healthy individuals. Patient studies remain important, because use of vasoactive medications, rapidly changing hemodynamics, hematomas of skull and extracranial tissues, and pulmonary dysfunction can all affect global and regional cerebral oxygenation. Therefore, it is important to exercise caution in translating our volunteer findings to patients. Our results reflect acute changes in Paco₂, which might change if the level is maintained for an extended period.

For this study, we focused on normocapnia and hypocapnia, and our results do not cover hypercapnic conditions often encountered in the ICU and operating room due to pulmonary insufficiency. Based on unpublished experience with targeted hypercapnia, we found that reliably maintaining a Paco₂ in the range of 50 mmHg with the addition of hypoxia was technically difficult due to issues with minute ventilation stability. In addition, the protocol necessary to encompass this third condition would be significantly longer. Therefore, we chose to focus on normocapnia and hypocapnia in this initial study. Elucidating the accuracy of cerebral oximeters under conditions of hypercapnia with varying FIO₂ remains an important question for future investigation.

In attempting to explain the variation observed in ScO₂ and oximeter bias, we used venous saturation and FOE as surrogates for CBF. As such, we made a number of assumptions about the expected changes in CBF and volume based on previous studies addressing the complex interplay of Pao₂, Paco₂, and cardiac output on cerebral hemodynamics. Because we did not measure CBF or blood volume directly using either transcranial Doppler or imaging techniques, it is impossible to quantitate to what degree our various experimental conditions altered these parameters. Future efforts combining measurement of CBF or cerebral blood volume with assessments of cerebral oximeter accuracy could provide insight into the basis of the observed changes.

Although the number of subjects was sufficient based on U.S. Food and Drug Administration recommendations for validation of oximetric devices, cerebral oximeter readings varied considerably between individuals, which increased with hypocapnia, leading to corresponding variability in the bias. This was especially true for the INVOS monitor, because the mean error of the measurements was higher than for the FORE-SIGHT. This variability may have prevented the resolution of trends in the bias by failing to reach significance under certain conditions.

Finally, different patient positioning may have an effect on accuracy. Although the 30° of head elevation in our study is similar to ICU positioning, a supine position is more common in the operating room. Although intracranial blood flow appears to be minimally affected by position changes based on transcranial Doppler³⁶  and thermodilution,³⁷  there is early evidence that ScO₂ signals may decrease with a change from supine to standing in stroke patients in the distribution of the stroke³⁸  and with head-up tilt in anesthetized patients,³⁹  although 20° changes in head position did not significantly affect cerebral oximetry in awake volunteers.²⁶  Unfortunately, the accuracy of cerebral oximeters has not been studied with varying position, and so it is unclear whether our study results reflect other positions.

The decreased CBF induced by hypocapnia results in a decreased ScO₂, with an increased bias. The oximeter is in effect overestimating the ScO₂ under this condition. This is relevant during measurement in patients undergoing induced hyperventilation. Understanding this increased bias is especially important for patients at risk for hypoxia-induced injury, such as in the setting of increased intracranial pressure.

Normalizing Paco₂ by eliminating hyperventilation and raising Pao₂ by increasing FIO₂ both increased cerebral oxygen saturation in our study. With hyperoxia, the ScO₂ reflected an increase in cerebral oxygenation but underestimated the actual amount. Future clinical studies verifying changes in cerebral oxygenation are essential to understanding whether recommended interventions of recent clinical trials using cerebral oximetry truly augment brain oxygenation. Our results demonstrate that changes in cerebral oximeter readings may not accurately reflect changes in cerebral oxygenation under hypocapnic conditions secondary to altered CBF.

Supported by paid studies at the University of California, San Francisco Hypoxia Research Laboratory (San Francisco, California) involving pulse oximeter performance evaluation not related to this research effort.

The authors declare no competing interests.