Analysis of high-frequency P_btO₂ measures in traumatic brain injury: insights into the treatment threshold

Abstract

OBJECTIVE

Brain tissue hypoxia is common after traumatic brain injury (TBI). Technology now exists that can detect brain hypoxia and guide corrective therapy. Current guidelines for the management of severe TBI recommend maintaining partial pressure of brain tissue oxygen (P_btO₂) > 15–20 mm Hg; however, uncertainty persists as to the optimal treatment threshold. The object of this study was to better inform the relationship between P_btO₂ values and outcome for patients with TBI.

METHODS

P_btO₂ measurements were prospectively and automatically collected every minute from consecutive patients admitted to the San Francisco General Hospital neurological ICU during a 6-year period. Mean P_btO₂ values in TBI patients as well as the proportion of P_btO₂ values below each of 75 thresholds between 0 mm Hg and 75 mm Hg over various epochs up to 30 days from the time of admission were analyzed. Patient outcomes were determined using the Glasgow Outcome Scale. The authors explored putative treatment thresholds by generating 675 separate receiver operating characteristic curves and 675 generalized linear models to examine each 1–mm Hg threshold for various epochs.

RESULTS

A total of 1,380,841 P_btO₂ values were recorded in 190 TBI patients. A high proportion of P_btO₂ measures were below 20 mm Hg irrespective of the examined epoch. Time below treatment thresholds was more strongly associated with outcome than mean P_btO₂. A treatment window was suggested: a threshold of 19 mm Hg most robustly distinguished patients by outcome, especially from days 3–5; however, benefit was suggested from maintaining values at least as high as 33 mm Hg.

CONCLUSIONS

This analysis of high-frequency physiological data substantially informs the relationship between P_btO₂ values and outcome. The results suggest a therapeutic window for P_btO₂ in TBI patients along with minimum and preferred P_btO₂ treatment thresholds, which may be examined in future studies. Traditional treatment thresholds that have the strongest association with outcome may not be optimal.

Brain hypoxia is a frequent and important consequence of traumatic brain injury (TBI).^1,7,36 Monitoring of brain hypoxia allows corrective intervention and ongoing titration of therapy. S_jvO₂ (jugular venous oxygen saturation) monitoring is an established monitoring technique that can intermittently assess the oxygen content of venous blood returning from the brain to provide insights into oxygen utilization in large portions of the cerebrum.²⁶ Approximately 2 decades ago, new technology emerged that allowed the invasive and continuous monitoring of smaller brain regions of interest.^18,33,35 The recently published BOOST II (Brain Oxygen Optimization in Severe Traumatic Brain Injury Phase II) randomized controlled trial of Licox technology (Integra) suggests that therapy guided by focal, continuous measurements of the partial pressure of oxygen in the brain (P_btO₂) significantly reduces brain hypoxia and also reduces mortality by approximately 10%.²⁴

Unfortunately, the treatment threshold for P_btO₂ has not been established with certainty. Although the third edition of the Brain Trauma Foundation’s (BTF’s) Guidelines for the Management of Severe Traumatic Brain Injury provided a Level III recommendation that P_btO₂ be maintained greater than 15 mm Hg, the fourth edition deemed that there is insufficient evidence meeting current standards to recommend a P_btO₂ threshold.^3,5 Many currently adhere to the Neurocritical Care Society’s consensus statement, which recommends maintenance of P_btO₂ > 20 mm Hg.¹⁵ Both of these sources acknowledge the weak nature of the evidence currently informing the treatment threshold.

Important advancement in the TBI field has come from analysis of large data sets, likely because these data sets help overcome the tremendous heterogeneity inherent to TBI patients.^11,20,25,29 Our group has a large, prospectively collected database of physiological measures from all patients admitted to the neurological ICU at San Francisco General Hospital during a 6-year period. This database also contains 1,380,841 P_btO₂ values recorded during every minute of the ICU stay of 190 consecutively treated patients with TBI. This large quantity of high-frequency data allows us to study the relationship between specific P_btO₂ values and outcome with high resolution and thus provide insight into what the optimal treatment threshold may be.

Methods

Patient Characteristics and Management

We studied a consecutive series of brain-injured patients who were admitted to the neurological ICU between 2005 and 2011 and underwent P_btO₂ monitoring. Patients were managed with a standardized treatment protocol strongly adherent to the BTF’s Guidelines for the Management of Severe Traumatic Brain Injury.^4,10 P_btO₂ monitoring over this time period was performed exclusively using Licox monitors (Integra). Over the duration of this study, most patients who had an external ventricular drain placed per BTF guideline recommendations also had a Licox monitor placed for P_btO₂ monitoring, although the decision to place a P_btO₂ monitor was based solely on the attending neurosurgeon’s judgment of medical need. These monitors were placed through a distinct burr hole ipsilateral to the external ventricular drain in a standardized location 2 cm lateral and 2 cm anterior to Kocher’s point. In most instances, the monitors were placed on the right side of the brain. The P_btO₂ threshold used in all studied patients was the contemporaneous 15–mm Hg threshold recommended by the third edition of the BTF guidelines.² Low P_btO₂ values were treated by means such as optimizing ventilator settings, investigating and treating pulmonary pathology, increasing the F_iO₂ (fraction of inspired oxygen), augmenting blood pressure, and optimizing intracranial pressure.

Data Collection

Using a computerized data acquisition system developed by our group in conjunction with Aristein Bioinformatics LLC, data were collected and stored automatically from patients’ bedside monitors in the ICU. Prospective data collection with this system adhered to patient privacy regulations and was approved by our institutional review board. Patient consent was not required due to the minimal risk to patients and lack of patient contact. The system recorded values at 1-minute intervals and initiated automatically once patient data appeared on a bedside monitor and continued for the duration of the ICU stay.

Data Curation

Data for each patient was extracted to separate Microsoft Excel (Microsoft Corp.) files maintained on an encrypted computer and analyzed using Matlab (Math-Works) computing software. Our group designed a program that calculated average P_btO₂ values between time periods specified by the analyst. The program also counted the number of epochs with P_btO₂ values below specified thresholds between specified periods. We chose to analyze brain tissue oxygenation values below 75 different thresholds from 0 mm Hg to 75 mm Hg in 1–mm Hg increments for various epochs post-injury. The proportion of P_btO₂ values that fell below each of the 1–mm Hg increment thresholds relative to the total recordings in the epoch was quantitated to account for variation in the number of observations between patients. This analysis is similar to one we recently performed to examine the relationship between mean arterial blood pressure values and outcome from spinal cord injury.^6,12

Data Analysis

Statistical analyses were performed using IBM SPSS (version 24, IBM Corp.) software. P_btO₂ data for proportions below each 1–mm Hg threshold increment were analyzed for each of the first 7 days post-TBI and for other epochs of interest. Data from patients with confirmed TBI, Glasgow Outcome Scale (GOS) score at discharge, and daily P_btO₂ data were included in the analyses. Receiver operating characteristic (ROC) curves were used to calculate the area under the curve (AUC) to test whether the number of recordings below each 1–mm Hg threshold could significantly predict mortality, represented as a binary outcome of GOS score 1 (dead) versus any of the higher GOS scores (alive) at discharge. Since we analyzed 75 separate AUCs for each of the 7 days, plus combined data for days 3–5 and days 1–7 post-TBI, a total of 675 separate AUCs were generated. In order to visualize a potential threshold, the p values from each of these AUCs for each threshold were plotted for each day post-TBI to determine at which threshold the prediction of mortality was nearing significance.

ROC curves tend to be a very conservative predictive statistic; therefore, a second wave of more liberal data-driven analyses was performed with the same criteria listed above using a generalized linear model (GLM). The GLM was used that was fit to a gamma distribution in order to best model the proportional data. The main effect of GOS score was tested for the proportion of P_btO₂ values below each of the 75 thresholds in 1–mm Hg increments for each day post-TBI, plus days 3–5 combined and 1–7 combined.

To assess for possible confounding in the univariate relationship between P_btO₂ values and outcome, we performed a multivariable analysis. We developed an optimal regression model that allowed us to benchmark the predictive validity of the 19–mm Hg P_btO₂ threshold from days 3 to 5 on GOS score at discharge against other putative predictors of outcome, including age, type of injury (focal or diffuse), and injury severity (Glasgow Coma Scale [GCS] scores from the field and emergency department).

Data Visualization

GraphPad Prism (GraphPad) and Microsoft Excel were used to plot graphs of the results, and multipanel figures were organized using Microsoft PowerPoint (Microsoft Corp.), Adobe Photoshop (Adobe Systems), and Illustrator CC 2017 (Adobe Systems). In all cases, error bars represent standard error of the mean.

Results

In our database, 190 patients were identified with TBI and recorded P_btO₂ values. A total of 1,380,841 P_btO₂ values obtained every minute were recorded and analyzed for these patients. For analyses stratified by outcome, our analysis was necessarily restricted to those patients with GOS scores available, which included 185 patients. This analysis was also restricted to values recorded in the first 30 days after ICU admission. Figure 1 demonstrates the inclusion criteria for this study.

FIG. 1.

Analysis of inclusion criteria and flowchart. Patients with P_btO₂ data and GOS scores at discharge with a confirmed TBI were included in the analyses. A first wave of conservative data-driven analyses was performed using ROC/AUC to test whether specific treatment thresholds predicted mortality (GOS score 1 vs > 1). A second wave of liberal data-driven analyses was performed using GLM to test the same hypothesis, followed by post hoc testing using the least significant difference (LSD) for thresholds with significant predictors on GOS scores.

The GOS scores used to stratify patients based on outcome reflect those that were available or could be calculated at the time of discharge. Because of low rates of follow-up inherent to trauma patients at our county hospital, a meaningful analysis stratified by outcomes at a later time post-discharge was not possible. Of the analyzed patients, 51 (27.6%) died (GOS score 1). Fifteen (8.1%) were in a persistent vegetative state (GOS score 2); 110 patients (59.4%) had a GOS score of 3 (severe disability), and 9 patients (4.9%) achieved moderate disability (GOS score 4).

Patient Demographics

Characteristics of studied patients stratified by outcome are presented in Table 1. TBI patients were most often male in all outcome groups. There was a significant difference in the age of patients in each outcome group (p < 0.0001). Those achieving the best outcome (GOS score 4) were also the youngest (31.7 years), while those who died (GOS score 1) had the oldest average age (56.4 years). Additionally, Table 1 shows that the average total hours monitored between outcome groups were significantly different (p < 0.0001). Those who remained in a persistent vegetative state had the highest average number of hours monitored (520.3 hours), followed by GOS score 3 (347.2 hours), GOS score 1 (236.3 hours), and GOS score 4 (127.0 hours).

TABLE 1.

Characteristics of 185 patients studied, stratified by outcome

	Died (GOS score 1)	Vegetative (GOS score 2)	Severe Disability (GOS score 3)	Moderate Disability (GOS score 4)	p Value
No. of patients	51	15	110	9
Male sex (%)	40 (78.4)	11 (73.3)	93 (84.5)	9 (100)	0.520
Mean age, yrs	56.4	36.1	37.0	31.7	<0.0001
Mean total time monitored, hrs	236.3	520.3	347.2	127.0	<0.0001
Mean total PbtO2 time monitored, hrs	128.8	149.9	126.7	53.4	0.117

Of all patients studied, 185 were recovering from TBI and had a GOS score available at the time of hospital discharge; characteristics of these 185 patients are stratified by outcome. ANOVA was used to analyze a difference between means while binomial logistic regression was used for categorical data.

Distribution of P_btO₂ Values

The distribution of P_btO₂ values at various time points for the 190 TBI patients can be seen in Fig. 2. The proportion of counts in 1–mm Hg P_btO₂ bins between 1 and 75 mm Hg are shown and demonstrate that although most values were above 15 mm Hg, a high proportion fell below this value. Of all recorded P_btO₂ values from TBI patients from days 1 to 30, 19.0% were below 15 mm Hg. In the first 7 days following ICU admission, 19.7% of values were below 15 mm Hg. From days 3 to 5, an epoch with prevalent brain swelling and secondary brain insults,³¹ 12.4% of TBI patient values were below 15 mm Hg.

FIG. 2.

Distribution of P_btO₂ values demonstrating that a high proportion of measures were consistently below treatment threshold throughout the first 30 days after ICU admission. P_btO₂ values were measured using the Licox system, and measures are plotted within 1–mm Hg ranges. A: Individual days. B: Multiday epochs. The vertical red line indicates a possible treatment threshold of 19 mm Hg. The blue line includes patients recovering specifically from TBI (n = 190). The indicated time is in reference to the time of ICU admission.

Relationship Between Average P_btO₂, Proportion of Measures Below P_btO₂ Thresholds, and Outcome

Figure 3A displays average P_btO₂ values subsequent to admission. Mean P_btO₂ values were above the current threshold of 20 mm Hg at every time point and in every patient group. Moreover, while each outcome group exhibited significantly different mean P_btO₂ values (p < 0.001), mean P_btO₂ values did not correlate well with outcome. The proportion of values below thresholds demonstrated a stronger correlation with outcome. A detailed analysis of different P_btO₂ thresholds found that 19 mm Hg most robustly discriminated patients by outcome, especially for the combined period of days 3–5 (and in particular for days 3 and 4). This is illustrated in Fig. 3B, where we plotted the proportion of P_btO₂ values below 19 mm Hg for groups stratified by GOS score at the time of discharge for various noncumulative epochs relative to ICU admission. This finding is important, as it suggests that the proportion of brain oxygenation below thresholds have a more important relationship with outcome than average P_btO₂ values.

FIG. 3.

Patient outcome is more strongly associated with the proportion of time spent below treatment thresholds than it is with mean P_btO₂ values. The proportion of P_btO₂ values < 19 mm Hg is associated with outcome from days 1 to 4 in TBI patients. A: Values were measured using the Licox system, and average P_btO₂ values are plotted in relation to time after ICU admission. B: The proportion of P_btO₂ values below a potential threshold of 19 mm Hg are plotted. The “n” used in statistical testing in (A) is the number of P_btO₂ measures, while the “n” in (B) is the number of patients. Mean values are shown. The error bars represent SEM. *Significant on analysis.

P_btO₂ Thresholds and Neurological Recovery After Head Injury

To further explore possible treatment thresholds for P_btO₂ management, we plotted the proportion of values below 75 different potential P_btO₂ thresholds for analysis in 1–mm Hg increments between 1 and 75 mm Hg (Fig. 4). When lines on the plot cross, diverge, or differ among GOS groups, a relationship between P_btO₂ and neurological outcome is suggested. Our identified threshold of 19 mm Hg is denoted by vertical red lines and tends to fall where values for the outcome groups are highly separated. Of interest, separation of outcome lines up to P_btO₂ values of approximately 45 mm Hg suggests that values up to 45 mm Hg are associated with outcome, particularly in moderate and severe disability outcome groups. This raises the possibility that there may be an outcome benefit to maintaining P_btO₂ values up to that threshold if they can be achieved.

FIG. 4.

The amount of time spent at lower P_btO₂ (< 45 mm Hg) values is associated with poorer neurological outcome after injury. Plots display thresholds at 1–mm Hg increments. All TBI patients with GOS values (n = 185) are reported. Values were measured using the Licox P_btO₂ monitoring system and time is relative to ICU admission. A: Individual days. B: Multiday epochs. The vertical red line at 19 mm Hg indicates counts below a proposed threshold of 19 mm Hg.

Data-Driven Discovery of a Candidate Therapeutic Tissue Oxygenation Window After TBI

Supplementing our qualitative analysis of the data, we deployed quantitative methodologies to hone in on P_btO₂ values that most strongly predict clinical outcome. The p values from predictive statistical models were plotted by the threshold being tested for predicting the GOS score at discharge to reveal any asymptotes or spikes in the predictive patterns that can help identify the most robust P_btO₂ threshold. Both the ROC curve (Fig. 5) and the GLM graphs (Fig. 6) revealed consistently low p values between 15 and 35 mm Hg. Post hoc testing for days on which significant results were identified (days 3–7) using least-significant-difference tests revealed the specific relationship between P_btO₂ thresholds at the peak of the p value dip for the GLM and GOS score at discharge, suggesting that 19 mm Hg may be a robust minimum threshold to maintain to prevent lower GOS scores at discharge. A value of 33 mm Hg emerged as a higher threshold that may afford greater safety and might be conceptualized as an ideal threshold.

FIG. 5.

Data-driven identification of potential treatment thresholds to predict mortality of TBI patients based on P_btO₂ proportions below each value using ROC analysis. P_btO₂ data for each TBI patient were analyzed using the ROC curve to calculate the AUC to test whether the proportion of events below each 1–mm Hg threshold could predict mortality (GOS score 1 vs GOS scores 2–4). p values from AUC results are plotted on the y-axis and 1–mm Hg thresholds on the x-axis to identify windows of vulnerability in the ICU for 7 days after admission to the ICU post-TBI (A–G). Patient numbers drop off each day, most likely due to discharge or death. The increase in patients on day 2 indicates that Licox monitoring was not initiated until 2 days after coming into the hospital for some patients. Data for all days, either 3–5 (H) or 1–7 (I), were combined to calculate proportions below each threshold for each duration, highlighting the average dip in p values that occurs between 15 and 40 mm Hg on days 3–5.

FIG. 6.

Data-driven identification of potential treatment thresholds to predict mortality of TBI patients based on P_btO₂ proportions below each value using GLM analysis. P_btO₂ data for each TBI patient were analyzed using GLM to test whether the proportion of events below each 1–mm Hg threshold could predict the GOS score at discharge (all groups combined/main effect). p values from GLM are plotted on the y-axis and 1–mm Hg thresholds on the x-axis, similar to Fig. 5, to identify windows of vulnerability in the ICU for 7 days after admission to the ICU post-TBI (A–G). Data for all days, either 3–5 (H) or 1–7 (I), were combined to calculate proportions below each threshold. The y-axis is plotted to emphasize the locations where the p values fall into significance (p < 0.05), with red arrows indicating the lowest point of the drop in p values.

Multivariable Analysis to Benchmark the Predictive Value of a P_btO₂ Threshold Against Other Variables

Our multivariable analysis served to benchmark the predictive validity of the suggested P_btO₂ threshold of 19 mm Hg from days 3 to 5 against possible confounding variables. In this model, age was the strongest predictor of outcome; the P_btO₂ threshold of 19 mm Hg was the only other significant predictor (Table 2). TBI type—defined as focal or diffuse—was not a statistically significant predictor of outcome in this model and neither was TBI severity as assessed by the GCS in both the field and in the emergency department.

TABLE 2.

Benchmarking validity of predictors on outcome—optimal regression

	Standardized Coefficients
	Beta	Estimate of SE	df	F	p Value
Age	−0.572	0.076	10	56.013	<0.001*
Proportion below PbtO2 of 19 mm Hg	−0.197	0.095	1	4.325	0.040*
TBI type (focal or diffuse)	0.027	0.070	1	0.150	0.699
GCS score in field	0.210	0.264	1	0.631	0.429
GCS score at ED	0.032	0.278	2	0.013	0.987

ED = emergency department.

An optimal regression was developed to benchmark the predictive validity of the 19 mm Hg P_btO₂ threshold from days 3 to 5 on GOS score at discharge against other putative predictors of outcome including age, type of injury (focal or diffuse), and injury severity (GCS scores from the field and emergency department). Age was the strongest predictor of outcome, and the P_btO₂ threshold of 19 mm Hg was the only other significant predictor (p = 0.040).

^*Statistically significant.

Discussion

Literature to date has demonstrated that low P_btO₂ values correlate with outcome and that correcting low values improves outcome.^{1,16,22,23,28,30} Much about brain oxygen monitoring remains to be refined, however. It is particularly important to define the optimal treatment threshold, and a very large data set of high-frequency P_btO₂ values from nearly 200 patients represents a unique opportunity to explore the relationship between highly resolved P_btO₂ values and outcome.

The Literature Basis for Existing P_btO₂ Treatment Thresholds

Several publications have outlined the relationship between P_btO₂ values and neurological outcome after head injury.^1,32,34 In 1998, Bardt et al. reported that patients spending > 30 minutes with P_btO₂ < 10 mm Hg had a significantly greater chance of unfavorable outcome and death than those spending < 30 minutes with P_btO₂ < 10 mm Hg. They concluded that low P_btO₂ values and duration of time spent at low P_btO₂ are associated with mortality; however, the study provided limited information regarding potential treatment thresholds.¹ Valadka et al. published a similar study in 1998 and also concluded that low P_btO₂ values and duration of time spent at low P_btO₂ are associated with mortality.³² More specifically, Valadka and associates concluded that the likelihood of death increased with an increasing duration of time below a P_btO₂ of 15 mm Hg or with any occurrence of values below 6 mm Hg. Van den Brink and colleagues also concluded in 2000 that low P_btO₂ values and the duration of time spent with low P_btO₂ are associated with mortality. They found that a 50% risk of death was associated with a P_btO₂ < 15 mm Hg lasting > 4 hours.³⁴ These past studies are in agreement in suggesting that lower P_btO₂ values and prolonged periods of hypoxia are associated with higher mortality rates. Together, the papers of Valadka and Van den Brink and their colleagues served as important evidence leading to the BTF’s published treatment threshold of 15 mm Hg in the third edition of the guidelines.³

Several works published more recently have also explored the effect of P_btO₂ values on outcome after head injury.^10,17,19 Maloney-Wilensky published a systematic review in 2009 of 3 studies including 150 patients recovering from severe TBI.¹⁹ It was concluded that brain hypoxia (P_btO₂ < 10 mm Hg) is associated with worse outcome after TBI and that treating patients to increase P_btO₂ may improve neurological outcome. In 2009 Chang et al. performed a retrospective analysis of hourly P_btO₂ measures in 27 patients who sustained severe TBI.⁷ They explored P_btO₂ cutoffs at 10 mm Hg, 15 mm Hg, 20 mm Hg, 25 mm Hg, and 30 mm Hg and found that the 20–mm Hg cutoff was most strongly associated with poor functional outcome. In 2010, Spiotta et al. reported improved outcomes in 70 patients in whom a P_btO₂ target > 20 mm Hg was maintained compared with historical controls.²⁸ In 2015 Lin et al. published a prospective randomized study to compare the effect of P_btO₂-guided therapy with traditional intracranial pressure–guided treatment on outcome in moderate and severe TBI patients.¹⁷ They employed a P_btO₂ threshold of 20 mm Hg and concluded that the survival rate in the P_btO₂-guided group was significantly increased at 3 and 6 months after injury, indicating that therapy directed by P_btO₂ monitoring is valuable for treatment in these patients. Eriksson et al. published a study in 2012 and concluded that the first 72 hours of P_btO₂ monitoring predicts mortality similar to the analysis we present in this study.¹⁰ They proposed that mortality is increased when the P_btO₂ monitor remains below 29 mm Hg in the first 72 hours of monitoring, raising the possibility that a higher threshold for brain oxygen monitoring should be considered.

The Neurocritical Care Society consensus statements on advanced neuromonitoring cited several lines of evidence in their recommendation to maintain P_btO₂ values above 20 mm Hg.¹⁵ They argue that the higher threshold provides a buffer that helps prevent critically low P_btO₂ values (< 10–15 mm Hg).²⁸ They also cited evidence that neuronal mitochondria have been found to require an intracellular pO₂ that corresponds to a P_btO₂ of about 20 mm Hg to maintain aerobic metabolism.^27,28 Additionally, P_btO₂ decreases below 20 mm Hg have been associated with other markers of cerebral ischemia or cellular dysfunction.¹⁵ We view our study as being largely confirmatory of the Neurocritical Care Society recommendations.

The recently published BOOST II randomized controlled trial employed the 20–mm Hg treatment threshold and was stopped early after meeting its predetermined endpoint, demonstrating that P_btO₂-guided therapy is not futile.²⁴ P_btO₂-directed therapy also demonstrated a 9% reduction in mortality in this study. Given this result, there is strong reason to further optimize brain oxygen–directed therapy and in particular the treatment threshold. The work we present here contains important, high-resolution insights into the relationship between specific P_btO₂ values, as we have analyzed—to our knowledge—the largest number of severe TBI patients with P_btO₂ values to date and over 3 times the number of P_btO₂ measures found in the next largest study.²⁸ Our analyses suggest that 19 mm Hg deserves consideration and further study as the minimum P_btO₂ treatment threshold. It additionally demonstrated that values at least as high as 33 mm Hg correlate with outcome, suggesting that there might be benefit to maintaining higher P_btO₂ values when they can be achieved.

Average P_btO₂ Values Correlate Poorly With Outcome

Remarkably, the results shown in Fig. 3A demonstrate that all average P_btO₂ values were above both published treatment thresholds.^3,15 In contrast, our analysis of the proportion of measures below thresholds demonstrated that a large proportion of values fell below the recommended thresholds. Simply looking at average values can thus be deceiving. Patient outcome correlated much more strongly with time spent below thresholds—in particular, 19 mm Hg—than with average values. Importantly, this suggests that efforts to avoid low P_btO₂ values are more important than those to augment them above threshold and that even brief periods of time spent below threshold may be harmful.

P_btO₂ Values Above 19 mm Hg and as High as 45 mm Hg May Be Beneficial

Our data suggest that a minimum P_btO₂ threshold of 19 mm Hg should be considered after head injury; this suggested threshold was derived in a traditional fashion based on the strongest statistical association with outcome. The importance of this P_btO₂ threshold is supported by our multivariable regression model, which served to control for putative variables confounding the relationship between P_btO₂ and outcome. In this multivariable analysis, the only other significant predictor of outcome besides age was the P_btO₂ threshold of 19 mm Hg from days 3 to 5.²¹

Importantly, our analysis also suggests that there may be therapeutic benefit to maintaining higher P_btO₂ values, as an association with outcome was robust up to 33 or 45 mm Hg, depending on the analysis performed (Figs. 4 and 6). Moreover, Fig. 4 demonstrates that higher P_btO₂ values seem more important for discriminating better outcomes while lower P_btO₂ values were more strongly associated with discriminating poorer outcomes. This indicates that patients might benefit from maintenance of P_btO₂ values above current thresholds.

With this in mind, our article is remarkable for suggesting distinct minimal and ideal treatment thresholds and thus the suggestion of a treatment window; currently, treatment thresholds in neurocritical care are conceptualized as single, discrete values. We contend that these may oversimplify the conceptual association between measured values and outcome. We must also caution that in striving for higher P_btO₂ values treating clinicians must take into account potential toxicities associated with higher oxygen values.^8,9,14 Thus, the potential physiological benefits of achieving higher P_btO₂ values must be balanced against the risks.

Limitations

A number of limitations must be considered in interpreting this study. Patient outcomes (GOS scores) were measured at the time of hospital discharge, as analysis at a later time point was not possible due to a high rate of loss to follow-up in this trauma population at our county hospital. Additionally, time was relative to ICU admission and not the time of injury. This time point is clinically relevant, however. This study does not inform the optimal intracerebral site at which to monitor P_btO₂ and does not allow analysis of the treatment threshold based on distance from focal brain lesions.¹³ Additionally, this study does not exclude the possibility that individual patients or some disease pathologies have a different relationship between specific P_btO₂ values and outcome. Ultimately, it will be necessary to validate candidate treatment thresholds prospectively. There are important limitations to such a study, however, as a limited number of thresholds can be studied concurrently, and it is unethical to randomize patients to thresholds believed to be suboptimal.

Conclusions

Although this study does not provide evidence for a causal relationship, it substantially informs the nature of the relationship between P_btO₂ values and neurological outcome. The proportion of measures below treatment threshold was more strongly associated with outcome than average P_btO₂. Moreover, our study supports the concept of a P_btO₂ treatment window in TBI recovery, with suggested minimum and ideal treatment thresholds. Indeed, our results suggest that a minimum P_btO₂ treatment threshold of 19 mm Hg may be considered and that values as high as 33 mm Hg or 45 mm Hg may be of additional benefit if they can be achieved. Validation of these results in a prospective study is needed.

ABBREVIATIONS

AUC

area under the curve

BTF

Brain Trauma Foundation

GCS

Glasgow Coma Scale

GLM

generalized linear model

GOS

Glasgow Outcome Scale

TBI

traumatic brain injury

P_btO₂

partial pressure of oxygen in brain tissue

ROC

receiver operating characteristic