Patients with severe traumatic brain injury (TBI) are frequently monitored using intracranial pressure (ICP) and brain tissue oxygen tension (PbrO2) catheters. Maintaining these variables in the normal range has been suggested to improve the clinical outcome of these patients [13]. Both of these methods are limited because they require an indwelling probe, are invasive, and can be associated with complications. In addition, many patients with low- or moderate-grade TBI are not monitored. Given that therapy following severe TBI is directed towards prevention of secondary brain insult [2], such as cerebral hypoxia, unmonitored patients may suffer undetected episodes of cerebral hypoxia that could lead to poor outcomes [4].

Near-infrared spectroscopy (NIRS), which is an attractive, noninvasive method for regional monitoring of cerebral oxygenation, has been used in cardiac surgery patients during intra- and postoperative periods. This method can result in earlier detection of cerebral complications, and its use has been associated with improved clinical outcomes in these patients [5]. NIRS is easy to use and provides continuous real-time assessment of regional cerebral oxygen saturation (rSO2). The NIRS apparatus, INVOS 5100 (NIRS monitor, Somanetics INVOS® system, Somanetics Inc., MI, USA), uses two wavelengths of near-infrared light and measures the ratio of oxyhemoglobin to total hemoglobin, which is presented as a percentage value, known as rSO2 [6].

Insufficient evidence currently exists to determine whether the information that rSO2 provides is useful for management or prognosis of TBI patients. Reports concerning the usefulness of NIRS are contradictory with respect to its technical reliability and clinical value. Absolute rSO2 values that are measured by cerebral oximetry can be affected by several factors. Patients may suffer forehead, skull, and cerebral frontal lobe lesions, which preclude the acquisition of a correct signal reading from the brain. rSO2 values are also affected by hemoglobin concentration, skull thickness, the differential path length factor, extracranial blood contamination, the area of the cerebrospinal fluid layer, and sensor location [7].

Importantly, NIRS monitoring has yet to be validated against an accepted measure of cerebral regional oxygenation. Some studies in TBI patients have documented poor correlation between jugular venous saturation (SjO2), which is a global measurement of cerebral oxygenation, and rSO2 [8], whereas others have found a similar response pattern for rSO2, SjO2, and PbrO2 after increment of fraction of inspired oxygen [9]. Dunham et al. [10] examined a small cohort that included 45 TBI patients and found that there was good correlation between rSO2 and other variables that are commonly measured in these patients, such as cerebral perfusion pressure (CPP) and PbrO2, as well as with clinical outcome.

Both NIRS and PbrO2 are measurements of regional cerebral oxygenation. Two short trials reported that dynamic changes in PbrO2 were reflected by changes in rSO2 [11, 12], but these trials included only a few flow-sheet-paired determinations. To the best of our knowledge, continuous time-domain analysis comparing PbrO2 with rSO2 has never been investigated.

We hypothesized that cerebral rSO2 would correlate with other variables that are commonly used in the management of TBI patients. A significant association between rSO2 and PbrO2, with the latter being considered the gold-standard value for the purpose of this study, would suggest that noninvasive cerebral oximetry provides relevant data. Therefore, the aim of this study is to investigate a continuous correlation between rSO2 and PbrO2 in 22 stable patients with severe TBI.


This prospective, observational, data-gathering study was approved by the Institutional Review Board of Hospital Universitario “Virgen del Rocío,” and informed consent was obtained from patients’ family and relatives. Consecutive subjects were enrolled over the course of 9 months, from July 2008 through to April 2009.

Patients with severe TBI [Glasgow Coma Scale (GCS) ≤9] and fulfilling all of the following criteria were included: (1) an intraparenchymal ICP/PbrO2 catheter had been previously inserted, (2) passing the initial resuscitation phase, and (3) being under controlled mechanical ventilation, with the patient sedated and fully adapted to the ventilator. Patients with frontal hematoma, wounds on either the scalp or the forehead, and those with deficient signals measured by NIRS or PbrO2 were also excluded from this study.

Study strategy

All patients were managed using TBI guidelines [13] and treatment algorithms that have been approved by our institution [13, 14]. A multiparametric 3 M monitor (GMS, Kiel-Mielkendorf, Germany) continuously recorded mean arterial pressure (MAP), CPP, PbrO2, and intracerebral temperature (TEMP). ICP and PbrO2 were monitored using the LICOX® IMC system (GMS). PbrO2 and ICP probes were inserted through a unique screw into the frontal lobe to monitor watershed territory between anterior and middle cerebral arteries of uninjured areas, according to computed tomography (CT) scan. A 120 min period prior to recording the oximeter measurements was used to establish equilibrium. Monitor placement was confirmed by manipulation of oxygen inspired fraction (FiO2) and CT scan.

The cerebral oximeter sensors were placed on both sides of the forehead with the medial edge of the sensor at the midline (Fig. 1). Cerebral oxygen saturation was monitored by near-infrared spectroscopy (INVOS 5100 Cerebral Oximeter; Somanetics Corporation, Troy, MI, USA). A 20 min period was taken in order to establish equilibrium prior to recording the oximeter measurements. “Ipsilateral rSO2” denotes rSO2 and PbrO2 readings from the same cerebral hemisphere.

Fig. 1
figure 1

a Self-adhesive skin patches, which contain a near-infrared light-emitting diode and two photodiode detectors to measure returning scattered light intensities of deoxyhemoglobin and oxyhemoglobin, were applied to the patient’s left and right forehead. b In order to assess regional transcranial oxygen saturation (rSO2), the detector near to the emitting diode measures photons in the superficial tissues (scalp and bone), whereas the far detector includes photons from the deep tissues (scalp, bone, and cerebral cortex). The signal difference between the near and far detectors allows calculation of regional capillary/venous oxygen saturation in the cerebral cortex. c Brain tissue oxygen tension probes, which use a polarographic technique to determine brain tissue oxygenation (PbrO2), were inserted through a unique screw (c) into the white matter of frontal lobe (d). rSO2 and PbrO2 were continuously and simultaneously measured in the same patient

Readings were collected after patient stabilization at the intensive care unit (ICU) and over the course of the first 16 h post injury. Two readings per minute were simultaneously collected for the following six variables: rSO2 (left and right frontal lobes), MAP, CPP, PbrO2, and TEMP. Thus, a complete paired data set was collected by the computer every minute and stored for later processing for statistical analysis.

Statistical analysis

Values are reported as means ± standard deviation (SD) for continuous variables and as percentages for categorical variables.

The associations among PbrO2 and rSO2, age, skull thickness, blood hemoglobin concentration, MAP, ICP, CPP, GCS, and TEMP were evaluated by multiple linear and logistic regression analyses.

Given that maintaining PbrO2 >15 mmHg remains common practice by clinicians, we created a multivariate logistic regression analysis model to assess predictors that patients would have PbrO2 >15 mmHg (dichotomous dependent variable). After examining correlations among variables, the independent contributions of age, GCS, hemoglobin concentration, skull thickness, intracranial temperature, CPP, ICP, and ipsilateral (with PbrO2) rSO2 were incorporated into our model. The model was constructed using a forward stepwise method, with the likelihood ratio test and model’s calibration assessed by the Hosmer–Lemeshow goodness-of-fit test. From logistic regression analysis, an adjusted odds ratio (AOR) and its 95% confidence interval (CI) were derived for each factor in the model. All the variables were introduced into the model after exclusion of the presence of collinearity, based on R 2 >0.7.

Finally, receiver-operator characteristic curve analysis was performed to evaluate the screening efficiency of ipsilateral rSO2 in predicting intracerebral hypoxia (PbrO2). We explored two PbrO2 cutoffs (≤12 and ≤15 mmHg). The sensitivity, specificity, predictive values, and likelihood ratios of optimal rSO2 cutoffs in predicting cerebral hypoxia were determined.

P value <0.05 was considered to be statistically significant. Data were analyzed using SPSS version 16.0 (Chicago, IL, licensed to Hospital “Virgen del Rocío”) for Windows.


Fifty-six patients with severe TBI were evaluated during the course of this 9 month study. Patients were excluded because of frontal hematoma (N = 10), wounds on either the scalp or the forehead (N = 9), PbrO2 catheter malfunction (N = 2), deficient NIRS signal (N = 6), refusal of informed consent by relatives (N = 2), and hemodynamic and/or respiratory instability (N = 5). A total of 22 consecutive TBI patients with mean age of 33 ± 13 (16–51) years and admission GCS of 6 ± 3 [39] were included in this study.

All patients had sustained blunt head injury, with most resulting in lesions compatible with diffuse injury. Subdural hemorrhage was described in four patients and subarachnoid hemorrhage in five patients. Most subjects had isolated TBI. The types of TBI sustained by the subjects, according to TCDB classification, are listed in Table 1.

Table 1 Patients’ characteristics (N = 22)

During 16 h of monitoring, a total of 42,240 paired records were obtained. Incomplete information, such as unavailable paired rSO2 and PbrO2 values, resulted in exclusion of 631 records from analysis. A total of 41,809 paired measurements were included in the final analysis. Mean ± SD of the records are summarized in Table 2. No measurement-related adverse reactions occurred, for either PbrO2 or rSO2.

Table 2 Total observations (records = 41,809)

A direct and adjusted [β coefficient 0.36 (0.35–0.37)] poor correlation between paired measurements of PbrO2 and ipsilateral rSO2 was demonstrated (Table 3). PbrO2 values were also directly and weakly related to CPP, GCS, and intracerebral temperature. By contrast, skull thickness and age maintained an inverse relationship with PbrO2.

Table 3 Multiple regression analysis of brain oxygen tension (PbrO2) with regional oxygen saturation (rSO2), cerebral perfusion pressure (CPP), intracranial temperature (TEMP), skull thickness, age, and Glasgow Coma Scale (GCS) (records: 41,809)

Regression modeling confirmed the independent association of ipsilateral rSO2 with cerebral oxygenation (PbrO2 ≥15 mmHg). The rSO2 increment was associated with an 11% increase in the probability of achieving PbrO2 >15 mmHg, which would reach the higher Wald value (the higher the value, the higher predictive value of the variable) (Table 4).

Table 4 Logistic regression analysis with PbrO2 <15 versus ≥15 mmHg as the dependent variable

Data were subsequently categorized with respect to clinically significant subgroups. The receiver-operating characteristic (ROC) curve demonstrated that rSO2 had low accuracy for detecting moderate (PbrO2 < 15 mmHg) intracerebral hypoxia [area under curve (AUC) with 95% confidence interval (95% CI): 0.6 (0.60–0.63); P < 0.0001]. For an optimal cutoff of rSO2 <70%, the positive predictive value (PPV) was only 5.4%, and the likelihood ratio for a positive test was moderately low at 1.2. Given that recent investigations [6, 10] have found that values of rSO2 <60% are well correlated with critical CPP and ICP, we repeated the calculations for this rSO2 cutoff and found that the sensitivity of rSO2 <60% for predicting moderate intracerebral hypoxia was significantly poorer (32%) (Table 5).

Table 5 Sensitivity, specificity, and positive (PPV) and negative (NPV) predictive values of the optimal rSO2 cutoff in differentiating patients with moderate (PbrO2 ≤15 mmHg) and severe (PbrO2 ≤12 mmHg) intracerebral hypoxia

By investigating the accuracy of rSO2 for predicting severe intracerebral hypoxia (PbrO2 <12 mmHg), we observed a moderate improvement in the AUC [0.82 (0.80–0.85); P < 0.0001] and sensitivity (73%), and an LR+ of 5.3 for a suboptimal cutoff of rSO2 <60% (Table 5).


The aim of this study is to determine whether NIRS correlated with an accepted standard for monitoring cerebral oxygenation, PbrO2. Although our results demonstrated statistically significant agreement between PbrO2 and rSO2 measurements, as assessed by linear and logistic regression, the ability of rSO2 to identify patients with intracerebral hypoxia was moderate. When the ROC curve was used to select an optimal cutoff for testing the accuracy of NIRS, it became clear that episodes of clinically significant focal cerebral hypoxia detected by PbrO2 were not detected by NIRS.

At best (cutoff PbrO2 ≤12 mmHg, Table 5), there was an unacceptably high rate of false-negative (27%) and false-positive results (14%), which could lead to untreated episodes of true cerebral hypoxia or unnecessary treatment of false cerebral hypoxia. Severe intracerebral hypoxia was better detected by NIRS than was moderate hypoxia (AUC = 0.80; LR+ = 5.3). Put another way, patients with severe intracerebral hypoxia were about 5.3 times more likely to have rSO2 ≤60% than were patients without intracerebral hypoxia. However, only LR+ values >10 mmHg have been considered to significantly increase the probability of having intracerebral hypoxia [15].

In contrast, we found that rSO2 had low accuracy for identifying patients with moderate hypoxia (AUC = 0.60; LR+ = 1.2). This low AUC value indicates low accuracy of rSO2 for predicting PbrO2 [16]. This phenomenon became evident when observing the results data depicted in Fig. 2, which show that the correlation worsens as the PbrO2 values increase.

Fig. 2
figure 2

Correlation between PbrO2 and ipsilateral rSO2, revealing poor correlation with wide limits of agreement

Detection of cerebral hypoxia is of paramount importance. Some observational studies have reported that low values of PbrO2 and SjO2 predict higher mortality rates and worse neurological outcomes in TBI patients [1721]. These results suggest that treatment based upon cerebral oxygenation variables may positively influence patient’s prognosis. Multiple episodes of cerebral hypoxia are often detected during the days that follow severe TBI, even when ICP and CPP are normal [22], which suggests the importance of continuously monitoring cerebral oxygenation. Unfortunately, not all TBI patients are monitored with PbrO2, and rSO2 could be an invaluable tool for selected patients when PbrO2 monitoring is not possible or is contraindicated.

In this study, we focused on the relationship between local oxygenation (PbrO2) and regional brain oxygenation (rSO2), which are two parameters of brain oxygenation that differ fundamentally in the type of data that they provide. PbrO2 has been proposed as a measure of extracellular tissue oxygen tension, thus reflecting the balance between oxygen delivery to the cerebral extracellular space and oxygen consumption by the cerebral tissue [23]. PbrO2 may be positively correlated with a wide range of parameters, including FiO2, PaO2, cerebral blood flow (CBF), fever, MAP, CPP [21], cerebral oxygen diffusion [24], and hemoglobin concentration [25], suggesting that it is influenced by perfusion and oxygen consumption. Many of these physiological variables were, in fact, correlated with PbrO2 in our study (Tables 3 and 4).

The parameters measured by NIRS are much less clear. NIRS interrogates arterial, venous, and capillary blood, and the derived measurements represent oxygen saturation in these three compartments [26]. Since most of the cerebral blood volume is venous [27], rSO2 may predominantly represent the saturation of hemoglobin in the venous blood bed. In fact, rSO2 may serve as a reliable indicator of changes in brain oxygenation induced by controlled hypoxemia in normal subjects [28], since it is directly correlated with FiO2 and PaO2 [29, 30]. By contrast, rSO2 may [31, 32] or may not [30] be related to CBF, CPP, and/or ICP. Similar to PbrO2, rSO2 has been suggested to represent a continuous measure of the balance between delivery and utilization of cerebral oxygen [26]. Thus, the fact that both variables were correlated in the multiple linear regression (Table 3) and logistic regression (Table 4) analyses is not surprising. In fact, rSO2 was the variable that was most strongly correlated with PbrO2, suggesting that each represents a method that can assess cerebral oxygen consumption.

Wide experience using NIRS as a predictive variable of clinical outcome has demonstrated decreased incidence of postoperative neurocognitive dysfunction in patients undergoing major surgery when NIRS was monitored and cerebral desaturation was treated [33]. By contrast, experience with neurocritical patients is contradictory. NIRS can monitor dynamic cerebral autoregulation in animal models [34] and septic patients [35] and has demonstrated a significant correlation with SjO2 in children [36]. Moreover, rSO2 may detect alterations in oxygenation before conventional pulse oximetry [37] and has been shown to have a direct correlation with clinical outcome [10]. However, NIRS and SjO2 values may not be correlated in TBI patients [8]. The usefulness of rSO2 readings as a screening tool for vasospasm or cerebral infarction after subarachnoid hemorrhage [30] or for detection of postoperative hematoma in patients with craniotomy [38] is limited.

Both PbrO2 and rSO2 have been concomitantly evaluated to investigate whether the two techniques monitor different dynamic changes of cerebral oxygenation in TBI patients. The significance level for the frequency of correlation was 90% [11], and spectral analysis demonstrated that PbrO2 and rSO2 signals contain similar information, even though they use completely different detection methodologies [12]. However, NIRS was not suitable as a part of neuromonitoring due to its high failure rate and limited sensitivity [39].

The present study has a number of limitations. First, although a significant association between PbrO2 and rSO2 was observed (Tables 3, 4 and Fig. 3), we recognize that the simultaneous comparison of these variables is problematic, as each monitor utilizes a distinct physical principle and measures a distinct physiological parameter. Second, rSO2 might partially reflect forehead oxygenation and may be affected by the thickness of the skull. Although we performed multivariate analysis to account for the thickness of the skull, we ignored the final influence of extracranial signals on rSO2 readings. Third, we must note that PbrO2 is not yet an accepted gold-standard technique for cerebral monitoring, although we used it as such in this study. Fourth, since the reference ranges of oxygenation parameters remain to be determined, there is no known normal rSO2 value. Fifth, we studied a highly selected sample of patients with severe TBI, and nearly 45% of the eligible patients were finally excluded from data analysis because of intra- and extracranial problems. Obviously, this could cast doubts on the applicability of NIRS in the whole TBI patient population. Last, although rSO2 was associated with PbrO2 >15 mmHg in the logistic regression model, Hosmer–Lemeshow analysis showed poor calibration (Table 4). A large number of data points (41,809 observations) with small, clinically insignificant differences in the predicted and observed outcomes should result in a significant Hosmer–Lemeshow goodness-of-fit test result [40].

Fig. 3
figure 3

Example of simultaneous recording of PbrO2 and rSO2 as real-time, continuous traces in a single patient. Time on the x axis spans 16 h. rSO2 increased and decreased in parallel with PbrO2. Right rSO2 ipsilateral with PbrO2

In contrast to previous studies that included data obtained from medical records, the strengths of this study include the relatively large number of data points collected and the thorough approach taken in utilizing prospective data that were collected simultaneously and continuously. As previously stated, this is one of the few studies comparing invasive versus noninvasive cerebral oxygenation in patients with severe TBI. Although both techniques provide different information, they could work together as a complementary multimodal monitoring technique to approximate cerebral oxygen supply and consumption, although this assumption remains largely speculative.

In conclusion, the data presented herein suggest that regional oxygen saturation measured by NIRS cannot precisely predict PbrO2 and, therefore, that measurement by NIRS should not be considered to be an acceptable substitute for PbrO2. Despite the fact that the use of several statistical approaches demonstrated significant direct correlation between PbrO2 and rSO2, noninvasive cerebral oxygenation as assessed by NIRS in severe TBI patients does not accurately reflect invasive cerebral oxygenation as determined by PbrO2. We cannot at present recommend systematic use of rSO2 as a substitute for PbrO2 in severe TBI patients.