Mortality and Prehospital Blood Pressure in Patients With Major Traumatic Brain Injury: Implications for the Hypotension Threshold

Key Points

Question  Is there a prehospital hypotension threshold for mortality in patients with major traumatic brain injury?

Findings  In this secondary analysis of the Excellence in Prehospital Injury Care Traumatic Brain Injury Study, the association between systolic blood pressure and adjusted probability of death was monotonic across a broad range (40-119 mm Hg), with each 10-point increase in systolic pressure associated with a decrease of 18.8% in the adjusted odds of death.

Meaning  In patients with traumatic brain injury, the concept that 90 mm Hg represents a unique or important physiological cut point may be wrong, and clinically meaningful hypotension may not be as low as current guidelines suggest.

Importance  Current prehospital traumatic brain injury guidelines use a systolic blood pressure threshold of less than 90 mm Hg for treating hypotension for individuals 10 years and older based on studies showing higher mortality when blood pressure drops below this level. However, the guidelines also acknowledge the weakness of the supporting evidence.

Objective  To evaluate whether any statistically supportable threshold between systolic pressure and mortality emerges from the data a priori, without assuming that a cut point exists.

Design, Setting, and Participants  Observational evaluation of a large prehospital database established as a part of the Excellence in Prehospital Injury Care Traumatic Brain Injury Study. Patients from the preimplementation cohort (January 2007 to March 2014) 10 years and older with moderate or severe traumatic brain injury (Barell Matrix Type 1 classification, International Classification of Diseases, Ninth Revision head region severity score of 3 or greater, and/or Abbreviated Injury Scale head-region severity score of 3 or greater) and a prehospital systolic pressure between 40 and 119 mm Hg were included. The generalized additive model and logistic regression were used to determine the association between systolic pressure and probability of death, adjusting for significant/important confounders.

Main Outcomes and Measures  The main outcome measure was in-hospital mortality.

Results  Among the 3844 included patients, 2565 (66.7%) were male, and the median (range) age was 35 (10-99) years. The model revealed a monotonically decreasing association between systolic pressure and adjusted probability of death across the entire range (ie, from 40 to 119 mm Hg). Each 10-point increase of systolic pressure was associated with a decrease in the adjusted odds of death of 18.8% (adjusted odds ratio, 0.812; 95% CI, 0.748-0.883). Thus, the adjusted odds of mortality increased as much for a drop from 110 to 100 mm Hg as for a drop from 90 to 80 mm Hg, and so on throughout the range.

Conclusions and Relevance  We found a linear association between lowest prehospital systolic blood pressure and severity-adjusted probability of mortality across an exceptionally wide range. There is no identifiable threshold or inflection point between 40 and 119 mm Hg. Thus, in patients with traumatic brain injury, the concept that 90 mm Hg represents a unique or important physiological cut point may be wrong. Furthermore, clinically meaningful hypotension may not be as low as current guidelines suggest. Randomized trials evaluating treatment levels significantly above 90 mm Hg are needed.

The societal burden of traumatic brain injury (TBI) is enormous; each year, TBI leads to 2.2 million emergency department visits, 280 000 hospitalizations, 52 000 deaths, and more than $60 billion in economic costs in the United States.^1,2 In addition, more than 5 million Americans have major long-term disabilities as a result of TBI.¹ Fortunately, there is growing evidence that proper and aggressive management of TBI in the minutes immediately following injury may improve patient outcomes by preventing or lessening secondary brain injury. This has led to the promulgation of evidence-based prehospital and in-hospital TBI treatment guidelines for both children and adults.^3-6

One major focus of these guidelines is the prevention and treatment of hypotension.^4,5 This is because it has been firmly established that even a single episode of hypotension during the prehospital or early hospital phases of TBI management is associated with dramatic increases in mortality.^3,7-26 Many studies have shown that low blood pressure (variously defined) increases the risk of death. However, the nearly universal assumption that a specific, clinically relevant threshold actually exists is entirely without support. In other words, the design of essentially every relevant study presumes a priori that there is a cut point below which outcome significantly worsens. However, simply dichotomizing small populations and then showing that it is worse to have lower blood pressure than higher blood pressure is not the same as identifying a true threshold. A clinically meaningful cut point would be one that correlates with a marked change in physiological response and patient outcome if blood pressure drops below that particular level. This requires study populations that are large enough to allow evaluation of blood pressure as a continuous variable rather than merely as a categorical variable, eg, low vs not low.

Given the absence of prehospital studies evaluating this specific issue, we analyzed the association between the lowest systolic blood pressure (SBP; obtained prior to hospital arrival) and mortality among children 10 years and older and adults in the Excellence in Prehospital Injury Care (EPIC) TBI Study.²⁷ Specifically, we tested the null hypothesis that no supportable inflection point in the relationship between SBP and mortality (ie, a threshold) would emerge from the data when evaluated without reference to any given definition for hypotension.

The parent study, EPIC, is evaluating the effect of implementing the prehospital TBI guidelines^3-6 for patients with major (ie, moderate or severe) TBI throughout Arizona. This is being done by using a before-after, multisystem, observational design. The study is expected to be completed in 2017 and has been described in detail elsewhere.²⁷ Rather than reiterating the details of the parent study here, we limit the description to the design attributes relevant to this specific secondary analysis. The patients in this evaluation are in the preimplementation cohort of the EPIC TBI Study. Postinterventional patients were excluded, since one of the emphases of guideline implementation is the prevention and aggressive treatment of hypotension. Thus, including these patients might introduce significant bias into this evaluation, as there was no intentional guideline implementation prior to the EPIC TBI Study.

The necessary regulatory approvals for the EPIC TBI Study have been obtained from the Arizona Department of Health Services and the State Attorney General. The University of Arizona Institutional Review Board and the Arizona Department of Health Services Human Subjects Review Board have approved the project and have determined that, by virtue of being a public health initiative, neither the interventions nor their evaluation constitute human subjects research and have waived informed consent and approved the publication of deidentified data.

The Arizona State Trauma Registry contains extensive trauma center data on all patients taken to the 8 designated level I trauma centers in the state. From the Arizona State Trauma Registry, all patients meeting study criteria were entered into the EPIC database. Each participating emergency medical services (EMS) agency then received a list of the patients in the EPIC TBI Study that were cared for in their system. The patients were matched by incident date, name, and other patient identifiers. Either scanned copies (paper-based patient care records [PCRs]) or electronic data files (electronic PCRs) were then sent to the study data center for entry into the EPIC database. This provided an extensive linked data set for study patients, which includes both prehospital and trauma center data. The entire process of identifying patients, linking EMS and trauma center data, accessing EMS PCRs, entering data, and structuring the EPIC database have been reported.²⁷ More than 20 000 patients have been enrolled in the EPIC TBI Study and more than 31 000 EMS PCRs have been entered into the database (patients cared for by multiple agencies have more than 1 PCR). The successful linkage rate is exceptionally high (eg, throughout the study, patients with missing data for SBP has been consistently less than 5%).

Inclusion criteria for the EPIC Study were physical trauma, a trauma center diagnosis(es) consistent with TBI (ie, either isolated or multisystem trauma that includes TBI), and at least one of the following definitions for moderate or severe TBI: Barell Matrix Type 1 classification, International Classification of Diseases, Ninth Revision head region severity score of 3 or greater, and/or Abbreviated Injury Scale head-region severity score of 3 or greater.²⁷

Exclusion criteria for this subgroup analysis included age younger than 10 years, an SBP less than 40 mm Hg or 120 mm Hg or greater, interhospital transfers, and death before arrival to the emergency department. In addition, patients that were missing data for age, SBP, or trauma type (ie, penetrating vs blunt) were excluded. The 120 mm Hg upper limit was chosen because this represents the highest reported threshold in the previous literature^{7-9,11,14,15,17-22,26,28-36} and because including a large number of patients with near-normal or normal perfusion in the mortality model would dilute the effects of the patients who are actually at risk for hypoperfusion.

This is a secondary analysis of the preimplementation cohort and entails no interventions.

The outcome is in-hospital mortality.²⁷

Continuous variables were summarized by median and range and were compared between the 2 cohorts (survived vs died) using the Wilcoxon rank sum test. Categorical variables were summarized by frequency and proportion (with 95% CIs) when appropriate and were compared between the 2 groups by Fisher exact test.

The overall trend in crude (unadjusted) mortality rates over the range of lowest prehospital SBP was explored using moving average plots. To plot the moving average, the crude death rate and corresponding 95% CI were calculated for patients with lowest SBP in each interval spanning 10 consecutive values (ie, 40-49 mm Hg, 41-50 mm Hg, 42-51 mm Hg, and so on, through 110-119 mm Hg). The estimated death rate and corresponding 95% CI were plotted against the midpoint of the interval (ie, the range of plotting is 44.5 mm Hg for 40-49 mm Hg, and so on, through 114.5 for the 110-119 mm Hg interval). The moving window of 10 mm Hg was selected to prevent any false cut points being created by data anomalies in the frequency of the last digit of lowest recorded SBP (eg, in the data set, even numbers were preferred to odd numbers, and the digit 0 was the most popular, followed by 8 and 6). Thus, using a window length of 10 prevents abnormalities arising from the uneven recording distribution of the last SBP digit.

The risk-adjusted associations between mortality and SBP were examined by logistic regression, which modeled the log odds of death, adjusting for important risk factors and potential confounders (ie, age, sex, race/ethnicity, payment source, trauma type, prehospital hypoxia, prehospital intubation, and treating trauma center). The linkage of EMS data to the Arizona State Trauma Registry allowed the use of actual diagnostic/anatomic injury scoring to adjust for overall injury severity (Injury Severity Score)³⁷ and TBI severity (International Classification of Diseases, Ninth Revision head injury diagnoses matched to Abbreviated Injury Scale head-region score)^38-44 rather than having to rely on far less reliable prehospital physiological injury indicators (eg, Glasgow Coma Scale score). The effects of continuous variables (ie, SBP and age) in the logistic regression were fitted nonparametrically using penalized thin plate regression splines through the generalized additive model.⁴⁵ The model was penalized to avoid overfitting (excessive “wiggliness” in the transformation function due to random noise), and the smoothing parameters were chosen to optimize the Akaike Information Criterion, a measure of the predictive power of the model.⁴⁵ Thus, the functional forms of these variables were determined by the data.

The software environment R was used for the analysis,⁴⁶ and the R package mgcv^45,47 was used for the generalized additive model. P values were calculated from a Wald-type test using the Bayesian covariance matrix.⁴⁸ All tests were 2-sided with α = .05.

There were 17 105 patients in the preintervention group from January 2007 to March 2014. Excluded were 1162 children (6.8%) younger than 10 years, 4823 (28.2%) interfacility transfers, and 6352 (37.1%) with a lowest prehospital SBP less than 40 mm Hg or 120 mm Hg or greater as well as 924 (5.4%) with missing data (SBP, 300; transfer status, 623; and trauma type, 1). This left 3844 patients (22.5%) in our study cohort.

Among these 3844 patients, 528 (13.7%) died. Table 1 summarizes the demographic information and patient characteristics by survival status. Figure 1 shows the crude (unadjusted) moving average of death rate by lowest EMS SBP. This plot reveals a relatively steady slope from 40 mm Hg to nearly 110 mm Hg. A logistic regression model was fitted that examined the effect of lowest prehospital SBP on mortality risk, controlling for risk adjusters and potential confounders. For continuous variables (ie, SBP and age), the functional form of the covariate effect was obtained nonparametrically with the value of the smoothing parameter calculated to optimize the Akaike Information Criterion. All other confounders were categorical (Table 1). Table 2 shows the effects and P values of all covariates in the model (except for the continuous variables and treating trauma center, which were all significant at P < .001). As has been found by many previous studies,^{7,8,11,17,18,49,50} hypoxia was a highly significant risk factor and was included as a confounder in the model. The data by trauma center, while parametric, are not shown in Figure 2. Because absolute anonymity is required by state regulations and the institutional review board (for patients, EMS agencies, and hospitals), we are not able to report specific trauma center–related data, even generically; because trauma center volumes are a matter of public record, presentation of these data could conceivably lead to hospital-specific information being inferred or identified (eg, because of comparisons of the sizes of the 95% CIs). However, because treating trauma center was a significant confounder, we adjusted for it in the model.

In the optimal model (based on Akaike Information Criterion), the adjusted effect of lowest SBP on log odds of death was nearly perfectly linear, with an adjusted odds ratio of 0.812 (95% CI, 0.748-0.883; P < .001) associated with a 10–mm Hg increase in SBP at any level between 40 and 120 mm Hg (eg, a patient with an SBP of 110 mm Hg has an 18.8% lower adjusted odds of death than one with an SBP of 100 mm Hg, and so on throughout the entire range). Figure 2 shows the adjusted probability of death over the range of 40 to 120 mm Hg. As can be seen, the rate of change in estimated probability of death is essentially constant. In other words, there is a striking absence of any identifiable threshold of SBP in relationship to mortality, and major reductions in both crude and adjusted mortality continue far to the right of the classic 90 mm Hg hypotension level. Additional evidence comes from the receiver operating characteristic curve plot of the data. The area under the curve is 0.705, and there is no cut point that gives satisfactory levels of both sensitivity and specificity to indicate a threshold.

The previous literature related to this investigation consists of studies that were small,^{7,8,11,14-21,23,24,26,29,30,34,50} had limited or no prehospital data,^{7,11,14-17,20,21,24,26,28,29,34,36,50} or evaluated general trauma populations (ie, were not specific to patients with TBI).^35,51-55 The current study is unique in both its size and its access to detailed prehospital data. A key reason for evaluating the effect of blood pressure measured before hospital arrival is because the injured brain is so highly sensitive to changes in perfusion, and the timeframe during which neuronal damage begins is so short. It is well established that secondary brain injury is initiated by even brief periods of compromised blood flow.^{4,5,11-13,17,20,27} Thus, decreased perfusion occurring during the prehospital time interval may have a profound effect on outcome. Indeed, our results reveal a strong, independent association between mortality and blood pressure measured in the field. This is remarkable, given the large number of factors that potentially affect survival in patients with TBI. It appears that the effectiveness of subsequent interventions may be highly dependent on patients who are neurologically viable being delivered to the trauma center so they have the potential to benefit from subsequent specialized care.

One of the most striking aspects of the literature evaluating the association between blood pressure and TBI mortality is the underlying assumption that there is a clinically relevant threshold. Some might argue that this is merely an operational reality inherent to the studies, that some level of hypotension must be chosen as a treatment threshold. However, even if the threshold concept isn’t always explicitly affirmed, its use is so ubiquitous that, functionally, it is treated as a given in the literature. In other words, there is a nearly universal concept of the existence of a level of SBP that represents a cut point, below which it is highly deleterious to drop.

However, the results of the current investigation seem to provide a significant contrast to current thinking about the implications of hypotension in the early care of patients with TBI. Visually evaluating the plot of adjusted mortality risk vs SBP (Figure 2) reveals a surprising finding—the absence of even a hint of a cut point at any level between 40 and 120 mm Hg. In addition, the mathematical expression of the data verifies this visual impression in that the association between SBP and the adjusted log odds of death is linear, with an adjusted odds ratio of 0.812 for mortality associated with a 10–mm Hg increase, regardless of the level being assessed. Thus, any 2 patients with an SBP difference of 10 mm Hg differ in their adjusted odds of death by 18.8%, which holds true across the entire SBP range. These results raise the possibility that, perhaps, no threshold exists in the sense that the concept is typically used. It appears that the threshold concept may have been artificially generated by investigations that, because of their small size, basically had no alternative but to deal with prehospital blood pressure dichotomously (ie, comparing low with not low). However, as this literature grew, the concept gained momentum and was incorporated into guidelines.

Another notable finding revealed by Figure 2 is the lack of a change in the slope even as the plot moves far to the right of the commonly applied definition for hypotension. This raises the possibility that clinically meaningful hypotension may not be as low as is currently thought for the injured brain. Indeed, despite the specifically recommended threshold, guidelines from the Brain Trauma Foundation also state that it is unclear what the threshold ought to be. Hence the explicit statement in the section on resuscitation end points: “The value of 90 mm Hg as a threshold for hypotension has been defined by blood pressure distributions for normal adults [emphasis added]. Thus, this is more a statistical than physiological finding.”⁵ Furthermore, the document goes on to forthrightly admit ambivalence about the recommended threshold: “Given the influence of cerebral perfusion pressure on outcome, it is possible that SBP higher than 90 mm Hg would be desirable during the prehospital and resuscitation phase, but no studies have been performed to corroborate this.”⁵ The lack of clarity surrounding this issue led the guideline authors to give it high priority in the section on “Key Issues for Future Investigation.” In the listing of recommended future research, the first topic is the identification of “the level of hypotension that correlates with poor outcome.”⁵

A careful reading of the extant studies reflects the complexity of defining hypotension in the setting of TBI. In fact, the literature varies widely and contains reports that have used cut points as low as 79 mm Hg and as high as 120 mm Hg in adults.^{7-9,11,14,15,17-22,26,28-36} Furthermore, the size and design of these studies preclude them from identifying “the” threshold, even if one actually exists. If previous prehospital studies had been larger, they would have been able to identify significant differences in outcomes using a wide range of potential thresholds, thereby revealing the arbitrary nature of choosing any one particular level.

To highlight this limitation in the current literature, we analyzed a broader cohort of patients in the EPIC database (SBP, 40-200 mm Hg) and dichotomized the cohort as “low” vs “not low” using various cut points in increments of 5 mm Hg. This yields the remarkable result that there is a statistically significant difference in the adjusted probability of death for thresholds as low as 60 mm Hg and as high as 135 mm Hg (Figure 3). In other words, one can pick any cut point throughout this range and obtain significant findings. Despite decades of assuming otherwise, it appears that the interaction between prehospital blood pressure and outcome may be physiologically continuous rather than dichotomous across a remarkably wide range. While it is hard to conceive of an approach to managing TBI that doesn’t include some level of blood pressure that requires treatment, it appears that the science that forms the basis for the current guidelines may require an entirely new way of thinking.

This study has limitations. First, the design is observational. Thus, we cannot establish cause and effect relationships associated with the treatment of hypotension. For instance, these data do not prove that the therapeutic target for blood pressure should be higher than the current recommendations. However, they do highlight the great importance of perfusing the injured brain and that blood pressure is powerfully linked to outcome.^16,25,28 Furthermore, these results do appear to support the statements in the TBI guidelines cautioning that the current recommendations may allow blood pressure to drop too low before intervening. A related concern is that we have not accounted for treatment of hypotension in the model. The parent study is designed specifically to identify the influence of treatment on outcomes using a controlled before-after system design, and the Analysis Plan²⁷ includes only an interim analysis (completed) and a final analysis (scheduled) and does not allow for multiple looks at the interventional data. Thus, to prevent any encroachment on the main study hypotheses, we are deferring all evaluations of treatment effects until the final analysis. Second, this evaluation does not inform questions associated with blood pressure management after the early resuscitative phase of care. This is true for several reasons; ongoing pressure monitoring in neurocritical care uses mean arterial pressure and cerebral perfusion pressure rather than SBP, and the prehospital management of blood pressure focuses solely on treating hypotension.⁴ Thus, the implications of our study cannot be used to inform issues associated with ongoing intensive care unit management or controversies, such as enhancing/optimizing perfusion.^56,57 Third, there were some missing data. However, for a prehospital study, the rate of missing data is extremely low (eg, 1.8% missing data for SBP; no missing data for mortality). Fourth, the database contains only those SBPs that were documented by EMS. Thus, we cannot know for sure that the reported measurements reflected the actual lowest SBP. Finally, there is no way to independently verify the accuracy of blood pressure measurements. However, this is true of essentially all EMS investigations.⁵⁸ One great advantage of the EPIC TBI Study is that the data team abstracts the PCRs directly and comprehensively. This level of scrutiny and consistency of data access is rare in prehospital research.⁵⁸

In a statewide, multisystem analysis of patients with major TBI, we found a linear association between the lowest prehospital SBP and the severity-adjusted probability of death across an exceptionally wide range. This suggests that there may not be a clinically meaningful threshold. Furthermore, for the injured brain, physiologically detrimental hypotension may occur at significantly higher levels than current guidelines suggest. These findings highlight the need for specific trials comparing various blood pressure treatment thresholds well above the classic 90 mm Hg.

Corresponding Author: Daniel W. Spaite, MD, Department of Emergency Medicine, University of Arizona College of Medicine, 1501 N Campbell Ave, Tucson, AZ 85724 (Dan@aemrc.arizona.edu).

Accepted for Publication: September 20, 2016.

Published Online: December 7, 2016. doi:10.1001/jamasurg.2016.4686

Author Contributions: Drs Spaite and Hu had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Concept and design: Spaite, Bobrow, Gaither, Denninghoff, Viscusi, Mullins.

Acquisition, analysis, or interpretation of data: Spaite, Hu, Bobrow, Chikani, Sherrill, Barnhart, Gaither, Denninghoff, Adelson.

Drafting of the manuscript: Spaite, Hu, Chikani.

Critical revision of the manuscript for important intellectual content: Spaite, Hu, Bobrow, Sherrill, Barnhart, Gaither, Denninghoff, Viscusi, Mullins, Adelson.

Statistical analysis: Hu, Chikani, Sherrill.

Obtained funding: Spaite, Bobrow, Denninghoff, Viscusi, Mullins.

Administrative, technical, or material support: Spaite, Bobrow, Barnhart, Gaither, Mullins.

Conflict of Interest Disclosures: Drs Spaite, Bobrow, Sherrill, Gaither, Denninghoff, Viscusi, and Adelson, Ms Chikani, and Mr Barnhart have received support from the National Institutes of Health via their university/academic appointments. No other disclosures were reported.

Funding/Support: Research reported in this article was supported by the National Institute of Neurological Disorders and Stroke of the National Institutes of Health under award R01NS071049.

Role of the Funder/Sponsor: The funder had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Previous Presentations: Presented in part to the National Association of EMS Physicians; January 16, 2014; Tucson, Arizona; and to the International Brain Injury Association; March 19, 2014; San Francisco, California.

Disclaimer: Research reported in this publication was supported by the National Institute of Neurological Disorders and Stroke of the National Institutes of Health under Award Number R01NS071049. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Additional Information: This is an observational, noninterventional analysis of a subset of the data in the Excellence in Prehospital Injury Care Traumatic Brain Injury Study. The parent study, while not a randomized clinical trial, is registered at ClinicalTrials.gov (NCT01339702).