Total Intravenous Anesthesia Including Ketamine versus  Volatile Gas Anesthesia for Combat-related Operative Traumatic Brain Injury

After approval from the institutional review board, Department of Clinical Investigation, Brooke Army Medical Center, San Antonio, Texas, we retrospectively reviewed the Joint Theater Trauma Registry for all patients presenting to one combat support hospital (CSH) treated between January 31, 2004, and May 1, 2005 (n = 4,882) (fig. 1).

Data included elements of the Joint Theater Trauma Registry as well as elements from inpatient and operative records such as mechanisms of injury, treatment at aide stations (level 1), forward resuscitative surgical suite (level 2), forward surgical teams (level 2), vital signs, and treatment at the CSH (level 3). Treatment and data from the CSH included type of surgery, duration of surgery, location of surgery, type of anesthesia, intraoperative anesthetic record, admission base deficit, arterial pH, serum glucose concentration, blood products administered during hospital course, intensive care unit duration of stay, ventilator days, number of days at level 3, and Injury Severity Score (ISS). For American casualties evacuated out of the theater of operations to Landstuhl Regional Medical Center (level 4) or military treatment facilities in the continental United States (level 5), data included ventilator days, intensive care unit days, and hospital days. The admission Glasgow Coma Scale (GCS) score was determined by using the best GCS score recorded from the scene or at level 1, 2, or 3. Craniectomy or craniotomy was performed at the discretion of the neurosurgeon based on type of skull injury, severity of injury, and intraoperative evidence of cerebral edema. Glasgow Outcome Score (GOS) using standard criteria ( appendix) was determined at the time of hospital discharge. A total of three neurosurgeons performed all procedures and had similar practices. Anesthesia care was delivered primarily by two providers, both with similar anesthetic goals. The goals included preventing secondary brain injury by decreasing ICP, cerebral edema, and CRMO²; maintaining hemodynamic stability and perfusion pressure; and avoiding hypoxemia and hyperglycemia. Providers, for example, specifically attempted to maintain hemoglobin greater than 10 g/dl, systolic blood pressure (SBP) greater than 90 mmHg, glucose less than 150 mmol/dl, partial pressure of arterial carbon dioxide (Paco²) 33–35 mmHg, and partial pressure of arterial oxygen (Pao²) greater than 100 mmHg; treat coagulopathy; minimize positive end-expiratory pressure; and maintain slight hypothermia (36°C) or normothermia.

In this study, two treatment groups for comparison included patients who received neurosurgical intervention with VGA or TIVA. These patients were not randomized to treatment or selected based on presentation. All patients presenting before August 21, 2004, received VGA. On August 21, 2004, neuroanesthesia providers started using TIVA techniques with anecdotal improvements in hemodynamic changes and operating conditions secondary to decreased cerebral edema. After August 21, 2004, 11 patients received VGA performed by anesthesia providers not on the neuroanesthesia team, whereas all others treated after this date received TIVA. As a result, treatment groups were nearly completely confounded by temporal period. Treatment selection was based solely on anesthesia provider preference and date of admission. The majority of patients received perioperative phenytoin, mannitol, and antibiotics. Balanced VGA included an opioid consisting of fentanyl or sufentanil and either sevoflurane or isoflurane. End-tidal volatile gas monitoring was not available. The TIVA techniques used propofol (75–150 μg kg⁻¹min⁻¹) combined with remifentanil (0.05–0.3 μg kg⁻¹min⁻¹), sufentanil (0.001–0.002 μg kg⁻¹min⁻¹), or fentanyl (0.01–0.02 μg kg⁻¹min⁻¹) and sometimes ketamine (5–20 μg kg⁻¹min⁻¹). Sodium thiopental was administered for cerebral edema at the discretion of the neurosurgeon. The primary outcome in our study was good (GOS 4–5) or bad (GOS 1–3) neurologic outcome at time of hospital discharge. In addition, subset analysis was performed on the TIVA group comparing those patients who received ketamine versus  no ketamine.

Demographic and clinical characteristics are presented as proportion for categorical variables, median (interquartile range) for nonparametric analysis using the Wilcoxon test, and mean ± SD for parametric analysis using analysis of variance. Groups were contrasted on binary outcomes with Pearson chi-square or Fisher exact test as appropriate. In an attempt to eliminate confounding by temporal period, treatment groups were contrasted on good neurologic outcome by application of the McNemar test22and matched-pair odds ratio after repeated one-to-one matching of patients receiving TIVA with those receiving VGA with regard to ISS (to within 10%), and perfectly on GCS (< 8, ≥ 8), base deficit (< 6, ≥ 6), Head Abbreviated Injury Score (≤ 3,> 3) and craniectomy versus  craniotomy. From these, subsets of approximately 30 matched pairs were selected without replacement 100 times, and the 100 matched-pair odd ratios and P  values relating treatment (TIVA, VGA) and good neurologic outcome (bad, good) were saved; the median and range of the 100 odds ratios and 100 P  values were reported. An unmatched treatment group contrast on good neurologic outcome was made with adjustment for a propensity score. Because treatment was not based on presurgical clinical characteristics, a modified propensity analysis was performed to adjust for injury severity. The variables we used to index injury severity were the craniectomy/craniotomy, hematocrit, and surgical time. A logistic model for treatment was applied in terms of these three variables. The propensity score, defined as the prediction probability from the logistic model, was used to adjust for injury severity in modeling good neurologic outcome from treatment, base deficit, and GCS. All statistical testing was two-sided, with a significance level of 5%, and SAS version 9.1 for Windows (SAS Institute, Cary, NC) was used throughout.

Review of the Joint Theater Trauma Registry revealed that 19% of patients (909/4,882) sustained traumatic head injury (fig. 1). Two hundred fifty-two TBI patients had neurosurgical interventions. Patients who underwent ICP monitor placement only (n = 4) or were not expected to survive the operation (n = 5) because of devastating injury determined intraoperatively by the attending neurosurgeon were excluded. Records were stored and maintained by the CSH patient administration, and after redeployment they were transported to the Patient Administration Systems and Biostatistics Activity (Fort Sam Houston, Texas), where they were subsequently scanned into PDF form. Complete records and follow-up data were unavailable for 12% (29/243). Twenty of the 29 patients (69%) had entire charts that could not be located, which limited our ability to meaningfully compare groups and outcomes. The remainder of missing data (9/29, 31%) included partial missing records and documentation of neurologic outcome, which was evenly distributed in both TIVA (4/9) and VGA (5/9) groups. Of the remaining 214 patients, 120 were anesthetized with VGA and 94 were anesthetized with TIVA. Demographic data are presented in table 1. We performed a nonmissing value analysis for each variable, and review of those data suggested that missing values were relatively balanced between the two groups among the subjects actually included in the analysis. Patients receiving VGA exhibited increased mean heart rate (± SD) (VGA: 93.5 ± 24.3 beats/min, TIVA: 83.8 ± 23.6 beats/min; P = 0.004), decreased mean hematocrit (VGA: 34.7 ± 7.6, TIVA: 38.3 ± 6.9; P < 0.001), and increased median international normalized ratio (VGA: 1.4, TIVA: 1.3; P = 0.02). Despite the statistical differences, groups seemed clinically similar for admission vital signs and laboratory data.

Table 2summarizes intraoperative data and postoperative outcomes between the two treatment groups, VGA versus  TIVA. Clinically and statistically significant differences included mean surgical time (VGA: 235 ± 98 min, TIVA: 182 ± 71 min; P < 0.001), mean amount of crystalloid administered (VGA: 2.7 ± 1.4 l, TIVA: 3.5 ± 1.5 l; P < 0.001), percentage with craniectomy (VGA: 60%, TIVA: 27%; P < 0.001), average hospital days (VGA = 12 and TIVA = 8; P = 0.04), mean GOS (VGA: 3.4 ± 1.4, TIVA: 4 ± 1.2; P = 0.001), good neurologic outcomes (GOS 4–5) (VGA: 54%, TIVA: 75%; P = 0.002), and risk of death (VGA: 16%, TIVA: 5%; P = 0.02) in VGA relative to TIVA. The discharge GOS was documented for VGA patients on average on day 10 (5–24) and for TIVA on day 7 (4–16) (P = 0.21). Although lowest SBP, change in SBP, and fresh frozen plasma use showed statistically significant differences between VGA and TIVA groups, none was thought to be clinically significant. In addition, the same analysis compared patients admitted before August 21, 2004, with those admitted on or after August 21, 2004. This is the date TIVA was first delivered. As expected because only 11 patients received VGA after August 21, the analysis mirrored the comparisons of VGA versus  TIVA.

Figure 2demonstrates the percentage of neurosurgical procedures performed by anatomical location. There were no differences (P = 1.0) in the percentage of right-sided (VGA: 52/88, 59%; TIVA: 36/88, 41%) or left-sided (VGA: 44/75, 59%; TIVA: 31/75, 41%) injury or procedures, although there were more bifrontal procedures in the TIVA group (24/40, 26%). Injury locations associated with the highest percent mortality were right occipital and left hemispheric.

A multivariate logistic regression model of good neurologic outcome (table 3), unadjusted for craniectomy versus  craniotomy, revealed that the odds of a good outcome were increased for subjects receiving TIVA (odds ratio, 2.32; 95% confidence interval, 0.99–5.45; P = 0.05), but the odds of a good outcome with TIVA were not significantly increased (odds ratio, 1.49; 95% confidence interval, 0.58–3.81) after additional adjustment for craniectomy versus  craniotomy and surgical time (table 4).

Treatment was confounded with temporal period; 90.8% (109/120) of patients receiving VGA were admitted before August 21, 2004, whereas 100% of patients receiving TIVA and 9.2% (11/120) of VGA patients were admitted on or after that date. Before August 21, 2004 (VGA group) was associated with markers of injury that may demonstrate increased severity of injury. These include increased mean heart rate (pre: 93.3 ± 25.2 beats/min, post: 85.2 ± 23 beats/min; P = 0.02), decreased hematocrit (pre: 35.1 ± 7.3, post: 37.5 ± 7.6; P = 0.02), increased operative time (pre: 235.3 ± 92.9 min, post: 186.9 ± 81.7 min; P < 0.001), decreased crystalloid (pre: 2.7 ± 1.4 l, post: 3.4 ± 1.6 l; P < 0.001), increased mean international normalized ratio (pre: 1.4 ± 0.3, post: 1.3 ± 0.3; P = 0.006), increased risk of bad GOS (1–3) (pre: 45.4%, post: 27.4%; P = 0.007), and increased risk of death (pre: 15.7%, post: 6.6%, P = 0.05).

Attempts to eliminate potential biases caused by confounding with temporal period were made by (1) a modified propensity score analysis and (2) matching on injury severity and four other variables. One hundred fifty-four patients (TIVA: 60, VGA: 94) had complete data for the five matching variables (ISS, GCS [< 8, ≥ 8], base deficit [< 6, ≥ 6], Head Abbreviated Injury Score [≤ 3, > 3], and craniectomy/craniotomy). After matching, 40 distinct matched pairs were found. From these, 30 matched pairs were selected without replacement 100 times. The median odds ratio relating treatment with neurologic outcome was 1.24 (range, 0.31–6.5), with a median P  value of 0.56 (range, 0.13–1.0). In a separate analysis, treatment was not associated with good neurologic outcome after adjustment for propensity score (odds ratio, 1.21; 95% confidence interval, 0.5–2.75; P = 0.65).

A subgroup analysis (table 5) was performed on the TIVA group to evaluate the effects of ketamine. No patients in the VGA group received ketamine. In the TIVA group, median ISS did not vary significantly with the use of ketamine (ketamine: 25, no ketamine: 25; P = 0.60), and ketamine use was not significantly associated with death (ketamine: 8.5%, no ketamine: 2.2%; P = 0.36) or good neurologic outcome (GOS 4–5) (ketamine: 79%, no ketamine: 72%; P = 0.47).

This retrospective study did not support our hypothesis. We could not demonstrate a significant difference in neurologic outcomes in patients with TBI requiring neurosurgical intervention when TIVA was compared with VGA. Specifically, when modified propensity scores were used to adjust for potential differences in injury severity and eliminate confounding factors such as admission hematocrit, SBP, and base deficit, which were associated with and predictors of neurologic outcomes and mortality, no differences were found. Conversely, TIVA including ketamine in a majority of patients seemed at least as efficacious and safe as VGA in these severely injured patients. This work supports the need and feasibility to perform prospective evaluations comparing TIVA versus  VGA in severe TBI. Although VGAs have been compared with propofol in animal head injury models, this study is the first human outcome study comparing VGA with TIVA for TBI requiring neurosurgical intervention.12–17 

The long history and reported safety of VGAs for neuroanesthesia in humans supports their use.11In addition, volatile gases are thought to be particularly useful for neurosurgical anesthesia because they decrease CMRO², attenuate surgical stimulation, provide neuroprotection, and allow rapid emergence. When administered prophylactically, recent studies suggest neuronal protection is conferred.23Although these properties support VGA use, they are also associated with many disadvantages, which drive efforts to discover superior anesthetics. Disadvantages associated with the VGAs include cerebral vasodilatation, increased CBF, increased cerebral blood volume, increased ICP, and cerebral steal syndrome.12–14VGAs also impair cerebral autoregulation in a dose-dependent fashion, creating increased dependence on CPP for blood flow.24,25As a result, decreases in mean arterial pressure (MAP) overwhelm vascular compensatory mechanisms and CBF decreases.26,27Finally, all volatile agents decrease systemic blood pressure and depress myocardial function in a dose-dependent manner, both of which exacerbate decreases in MAP and subsequently lower CPP.28 

Many experts recommend a propofol-based TIVA for neuroanesthesia because many of the adverse effects associated with VGAs are eliminated.15In a rat model of TBI with moderate hypothermia, propofol preserved ICP, CPP, and MAP better than isoflurane.29In addition, propofol was found to have neuroprotective properties.16,30–32Similar to the VGAs, propofol decreases CMRO², but unlike the VGAs, propofol also preserves CBF, produces cerebral vasoconstriction reducing cerebral blood volume, and reduces ICP.27,33Cerebral blood volume reduction in neurosurgery can be one of the most neuroprotective measures available; it establishes a slack brain for optimum operative conditions. Propofol also preserves cerebral autoregulation allowing compensation for reduced or increased MAP.13Propofol, however, like VGAs, reduces systemic vascular resistance. Decreases in systemic vascular resistance are particularly troublesome in hypovolemic bleeding patients seen in combat-related TBI. This effect is potentiated with propofol in hypovolemic patients.34 

Ketamine was used for several reasons, including the perceived benefit of improved hemodynamic stability in this complex patient population. The addition of ketamine to a propofol TIVA can attenuate systemic hypotension because its sympathomimetic effects preserve MAP.35–37Historically, though, ketamine is thought to be contraindicated for neuroanesthesia because of adverse cerebral vascular and neurologic effects.38,39Ketamine is reported to increase ICP, increase cerebral blood volume and flow, be epileptogenic, and increase CMRO². Recent research disputes these long-held beliefs.18,40In older studies, ventilation was not controlled; ketamine actually decreases minute ventilation in spontaneous breathing nonintubated patients. This decrease in minute ventilation resulted in expected increases in Paco²and subsequent CBF increases, cerebral blood volume increases, and elevated ICP. When ventilation is controlled, however, ketamine, especially when combined with propofol or midazolam, tends to decrease ICP and CMRO²and not increase them.19,41Positron emission tomography scanning after ketamine administration demonstrates little change in CMRO²and only marginal increases in cerebral blood volume.20Regional oxygen extraction fraction decreases with ketamine alone.20Furthermore, ketamine possesses anticonvulsant properties in normocapnic patients at lower doses and does not seem to be epileptogenic as previously thought.42,43 

Despite the favorable increases in SBP and improvements in hemodynamics associated with ketamine, subgroup analysis in our study demonstrates that ketamine was not associated with improved neurologic outcome. The low doses used in our study did not confer clinically significant differences in SBP compared with the VGA group. Conversely, ketamine was not associated with worsened neurologic outcome or mortality. At the doses used, we did not appreciate any psychomimetic phenomenon. Although the use of a propofol-based TIVA is supported by this study, the role of ketamine for neuroanesthesia remains unclear.

Beyond the hemodynamic and cerebral vascular effects, the potential role of anesthetics to effectively modulate or attenuate secondary brain injury in humans has received little attention, including at a recent National Institutes of Health workshop that comprehensively evaluated the evolution of prevention and treatment of secondary brain injury over the last few decades.8Recent research in TBI, however, targeting cellular interactions, offers more promising strategies to improve neurologic outcomes.9,10Examples of evolving areas of research include reduction in scavenging oxygen free radicals, prevention of excitotoxicity, maintenance of electrochemical ionic gradients, and inhibiting apoptotic mechanisms.9,10The unique contribution of anesthetic techniques to interact with these cellular targets is also evolving. Unfortunately, clinical studies documenting outcomes associated with differing anesthesia techniques are lacking.11 

One of the proposed mechanisms of anesthetic techniques on cellular targets is modulation of the antiinflammatory and proinflammatory cascade.30–32,44,45In fact, propofol is associated with antioxidant effects, reducing oxygen free radicals and further cell injury as well as decreased infarct volume size in cerebral ischemia animal models.17,31,32Other potential mechanisms of propofol’s action include γ-aminobutyric acid receptor inhibition and decreases in striatal dopamine accumulation after neuronal damage.46Finally, propofol prevents excitotoxic glutamate injury during neuronal ischemic events.16 

Unique to our study is the population of combat-related neurotrauma. These patients present complex problems for the anesthesia provider choosing anesthetic drugs. Challenges include the presence of polytrauma in 30% of patients, severe hemorrhage, and long evacuation times.6Despite these challenges, mortality for patients requiring neurosurgical interventions for TBI has decreased from 19% during the Lebanon conflict, to 16% during Vietnam, to 11% during Operation Iraqi Freedom.6,7,47,48This mortality difference is likely multifactorial, including improved trauma care and improved neurosurgical techniques.7,49 

The findings of decreased need for craniectomy and operative surgical time in the TIVA group could be confounded by time, neurosurgeon experience, or severity of injury, but they may also represent treatment effect. One hypothesis is that clinically relevant changes, such as less cerebral edema, may have resulted in these patients from TIVA, compared with those receiving VGA. Unfortunately, the neurosurgeons were not queried at the time of operation to support this contention.

There are important limitations in this study, especially the use of retrospective data. Unfortunately, the data and follow-up are limited because the data were obtained during active war. Temporal period, before and after August 21, 2004, was associated with indicators of injury severity and was confounded with treatment group (VGA, TIVA), rendering most treatment group contrasts potentially biased. Further, many other confounding factors, such as medical lessons learned resulting in improved processes and care over time, improvements in care and time during air evacuation, and increased rotations of medical providers, could not be controlled. However, there is a chance that differences between the VGA and TIVA groups did exist. Three of the 10 admission variables (heart rate, hematocrit, international normalized ratio) when stratified by group or date were statistically significant and demonstrated that the VGA patients may have presented to the CSH more severely injured. However, these three variables were likely not clinically significant, although we cannot exclude their importance. If patients were more severely injured in the VGA group, we would have expected to see more consistent differences in all of the admission variables in the stratified groups. Some other differences included a slightly higher incidence of bifrontal injury in the TIVA group versus  the VGA group. Conversely, the admission GCS difference between groups was not statistically significant but was found to be independently associated with good or bad outcome. Moreover, Head Abbreviated Injury Score was not different between the VGA and TIVA groups. Therefore, the small difference in GCS between groups could impact outcomes in favor of TIVA. The effect of polytrauma in approximately 30% of our patients on neurologic outcome could also confound our analysis. As a result, we chose ISS, which is an assessment of total body injury over Abbreviated Injury Score to control for severity of injury. Several patients in both groups had surgeries for other injuries, and this was not evaluated in the analysis. Unfortunately, we were not able to obtain GOS determined 6 months after injury. In addition, logistical changes, rotations of anesthesia providers, individual provider preferences, access to records, and limitations imposed on research during military operations precluded the continued accrual of patients to improve the power needed to determine differences in anesthetic techniques.

The National Institutes of Health conference held in 2000 addressed study design issues in TBI and recommended a 6-month follow-up time point.8Experts also recommended the use of the extended GOS, which provides a better distinction between levels of disability and correlates well with several other outcome measurements after TBI, including neuropsychological and cognitive testing, disability rating score, and measures of perception of health (36-item Short-Form Health Survey).50–53Applying the extended GOS and application of detailed neurocognitive, behavioral, and psychological testing was not possible in the combat environment.

The numerical differences in mean hospital days, intensive care unit days, and ventilator days with regard to treatment temporal period are unfortunately difficult to interpret. These differences may be secondary to changing logistical constraints, casualty loads, and discharge practices at the CSH. For Americans air evacuated from Iraq to Germany and the continental United States, the effect of transport and changing levels of providers could not be controlled. In addition, while between-group differences in fluid resuscitation, blood product administration, and vasopressor use were not perceived as clinically significant, they could not be controlled for. These interventions could change the findings of this study.

While mortality is commonly a primary outcome in comparative studies, neurologic outcome is an equally acceptable outcome for TBI. As such, TIVA use did not reach statistical significance in multivariate analysis of mortality, but it is likely that this analysis was underpowered to detect such a difference. Combining mortality with neurologic outcome improved the power of our analysis. Moreover, these results may not be applicable to non–combat-related TBI. Finally, this study is hypothesis generating rather than conclusive. Based on our data, a prospectively planned study would require approximately 90 subjects per group to achieve a power of 80% to detect a benefit from TIVA relative to VGA with regard to good neurologic outcome equal to that presented in table 2, assuming two-sided testing with a significance level of 5%.

In conclusion, this study evaluating GOS at hospital discharge for combat-related TBI patients requiring operative neurosurgical intervention could not determine differences between propofol-based TIVA, often including ketamine, compared with VGA techniques. Confounding variables associated with injury severity and temporal periods significantly limited our ability to make definitive conclusions regarding the outcomes we studied. Unfortunately, the nature of combat research limited our ability to increase the numbers of patients studied or to prospectively randomize patients so we could appropriately power our study. Despite the retrospective design of this study and the concern that patients who received TIVA may have been less severely injured, this is the largest cohort of combat-related TBI patients reported to date. Given the potential widespread effects of the improvements in neurologic outcomes of this anesthetic strategy, we believe that further prospective evaluation is needed to determine whether outcome-related differences exist.

The authors thank the 31st Combat Support Hospital Research Group, as well as Susan West, B.S.N., C.N.O.R. (Data Registrar), Monica Dewitt, B.S.N. (Trauma Coordinator), and COL John B. Holcomb, M.D. (Commander), from the United States Army Institute of Surgical Research, San Antonio, Texas, and Brooke Army Medical Center, Department of Clinical Investigation, San Antonio, Texas; National Capital Consortium Neurosurgery Residents, Washington, D.C./Bethesda, Maryland; LTC Jefferey Poffenbarger, M.D. (Neurosurgeon), Department of Surgery, Brooke Army Medical Center, San Antonio, Texas; LTC Rocco Armondo, M.D. (Neurosurgeon), Department of Surgery, Walter Reed Army Medical Center, Washington, D.C.; Lee Ann Zarzabal, M.S. (Biostatistician), Joel E. Michalek, Ph.D. (Professor), and John E. Cornell, Ph.D. (Professor), from the Department of Epidemiology and Biostatistics, University of Texas Health Science Center, San Antonio, Texas.