Introduction and Methology

INTRODUCTION AND METHODOLOGY

Part 2 of this document presents early clinical indicators that may be prognostic of outcome among patients with penetrating brain injury (PBI). This portion of the document is derived from the same principles of clinical epidemiology described in Part 1 of this Guideline's methodology section, and evaluates the pertinent literature on prognosis in patients with PBI qualitatively. The clinical factors used to determine prognosis for this section of the Guideline were determined by clinical assessments derived from measures with proven reliability (e.g., pupillary reflex or Glasgow Coma Scale [GCS] score). In this context, reliability means that different people with different backgrounds will reach the same conclusion about what they observe most of the time. Fortunately, good studies testing the reliability of these measures have been carried out, and will be discussed in the section on assessment that follows. Only reliable measures were included in this document. These include the following: clinical assessments (GCS score, pupillary size and reactivity, intracranial pressure [ICP], and neuroimaging), factual data (demographics, e.g., age), epidemiology (cause and mode of injury and caliber of weapon), and systemic measures (hypotension, coagulopathy, respiratory depression). Demographic factors such as race and gender were excluded because the literature lacked data on these topics. If we liken clinical assessments to diagnostic tests, and particularly to their role as predictors of poor outcome, we must be able to determine whether the diagnostic test has sensitivity, specificity, and positive or negative predictive value. An explanation of these factors is described below in a bayesian contingency table (Fig. 1).

Fig. 1

In a diagnostic test, sensitivity and specificity measure the appropriateness of the test. However, in prognosis, the most important aspects of this model are positive and negative predictive value. Positive predictive value represents the number of patients who had the clinical sign or prognostic indicator and had a poor outcome. For this to be meaningful and useful, a minimum positive predictive value of 70% or greater is thought to be necessary. However, this is inappropriate in PBI, since the prevalence of death is so high. Therefore, a different and more precise measure of prediction is used, as discussed below.

To analyze prognostic parameters in nonpenetrating traumatic brain injury (TBI), a probability of at least 70% of a given outcome has been used as a cut-off point in defining parameters of clinical importance. ¹ According to the bayesian statistical approach, the probability of a defined outcome in analyzing prognostic features should be viewed with respect to the 'prior probability, i.e., the distribution of outcome in the population studied. Since the prevalence of death (the primary outcome measure) in PBI is already above 70% in most series, when assessing positive predictive values of various prognostic indicators, a cut-off point of 70% is not relevant to this population. Therefore, instead of using a high positive predictive value as the standard of clinical importance, we used the odds ratio.

The odds ratio is an estimate of relative risk for the outcome of interest if the studied factor is present as opposed to the factor being absent. Data from each article were used to determine the odds ratio for each prognostic indicator. For example, an odds ratio of 1 indicates no increase in risk if the factor is present, and an odds ratio of 3 indicates a three times greater risk of the same outcome. A 95% confidence interval was also calculated for each odds ratio. Clinical significance was defined as an odds ratio whose 95% confidence interval did not include the number 1.

Prognosis studies (including prognosis with treatment) can have strength or weakness just like studies of therapeutic effectiveness. In the strongest studies, the patients should:

* Be seen at a uniform time in their disease (e.g., at time of admission to treating facility).

* Be followed prospectively for a designated block of time (e.g., for the first 24 hours after injury).

* Have their outcomes measured definitively and reliably (e.g., death).

* Be part of a continuous or defined cohort of at least 25 patients (e.g., from an ongoing, prospectively collected database).

* Be examined for extraneous prognostic variables, such as underlying disease or age (e.g., use of appropriate statistics such as multivariate analysis).

Published studies that report on prognosis cannot be assessed using the same paradigm as those for therapeutic effectiveness. Prognosis studies are observational in nature and can never be randomized controlled trials, which are meant to compare treatments. However, for consistency in assessing levels of evidence in reaching conclusions about prognosis, the same designations (Class I, II, and III) as those for therapeutic effectiveness were identified. The Guideline development group agreed on the following definitions:

* Class I-articles with all of the above characteristics.

* Class II-articles exhibiting four out of the five characteristics (including prospectively collected data).

* Class III-articles with three or fewer of any of the above characteristics.

Using this classification scheme, all published articles were evaluated and are listed in the evidentiary tables within each Guideline section. It should be noted that an observational study, such as a case series that might be Class I by the above criteria, would only be Class III if it were being used as an article on therapeutic effectiveness. These designations are meant to indicate scientific rigor and, thus, strength of evidence. However, unlike therapeutic effectiveness, articles on prognosis cannot be transposed from classification to recommendation because recommendations concern treatment, not prognosis. In the Guideline sections on assessment that use prognosis studies, therefore, the authors reached conclusions rather than made recommendations.

Outcome Distribution and Mortality in Penetrating Brain Injury: Implications for Prognostic Analysis

In prognostic analysis, features are analyzed with respect to defined outcome categories. Outcome in nonpenetrating and penetrating brain injury series is generally reported according to the Glasgow Outcome Scale (GOS). ² To analyze prognostic features of non-PBI, the GOS is commonly dichotomized into unfavorable (dead/vegetative/severe disability) versus favorable (moderate disability/good recovery) outcome. The outcome distribution in series of nonpenetrating TBI is typically U-shaped, ³ implying an inherent dichotomy in outcome consistent with the differentiation into unfavorable versus favorable (Fig. 2A). However, in PBI, the outcome distribution differs significantly. As is shown below, an overwhelming majority of patients have an unfavorable outcome; mostly, death. Thus, the appropriate outcome measure for PBI is mortality, not GOS score. It should be noted that the majority of studies addressed here deal with patients admitted to the hospital and do not account for prehospital deaths.

Fig. 2

The outcome distribution for PBI, shown in Figure 2A, is derived from a meta-analysis of results reported from five unselected series of civilian gunshot wounds. ^4-8 Full data are given in the evidentiary tables. A large number of deaths and low incidence of patients in the categories vegetative or severe disability characterize the figure. Consequently, the shape of the curve is similar to that of a 'hockey stick. In contrast, the outcome distribution in nonpenetrating brain injury is more typically U-shaped (Fig. 2B). However, among survivors, the percentage of patients with favorable outcome (moderate disability plus good recovery) is equal: 74%. Consequently, the main difference in outcome between penetrating and nonpenetrating brain injury is the difference in mortality.

Series of PBI that focus on patients with a low GCS score show much higher mortality rates (Table 1, selected series ^4,6-12) than for nonpenetrating TBI. Series that include patients dead on arrival or dead at the scene of the accident show mortality up to 93% (Table 2). ^4-7,9,13-23 Mortality rates in unselected series that include a wide range of GCS scores are shown in Table 2 with rates varying between 34% and 79%, explained by inclusion of patients without dural penetration, ¹¹ high number of 'mild injuries (GCS score of 13-15), ¹⁵ or low percentage of suicide as a cause of injury. ¹⁸ The rate of mortality and unfavorable outcome is considerably lower in nonpenetrating TBI. The most appropriate comparison of outcome can be made between patients with PBI versus nonpenetrating TBI from the Traumatic Coma Data Bank. In PBI, ¹⁰ overall mortality is 88% and the percentage of unfavorable outcome is 97% at discharge as compared with 32.5% mortality and 74.5% favorable outcome in nonpenetrating TBI at discharge. ²⁴ Because of the high death rate in PBI and relatively low number of patients in the other bad outcome groups, in contrast to nonpenetrating PBI, death is a more appropriate outcome measure for prognostic analysis than the dichotomized GOS.

Table 1

Table 2

Timing of Outcome Measure in PBI

Implications for Classification of Data

Death after PBI usually occurs soon after injury. Cavalieri et al. ²² describe 95% of deaths occurring within 3 hours and 97% within 24 hours of injury. Siccardi et al. report 92% of deaths occurring within 3 hours. In the study reported by Grahm et al., ⁶ 77% of patients died before admission to the intensive care unit; in the study by Shaffrey et al., ¹⁹ 53%. Hernesniemi ⁵ describes 76% of all deaths occurring in the first 24 hours, and Jacobs et al. ⁹ also found that 93% of deaths occurred in the first 24 hours. The percentages of death occurring within 12, 24, or 48 hours after injury, as reported in three studies, are shown in Table 3. The observation that, of the patients who die, the majority of deaths occur within the first 24 hours (70%) implies that the majority of outcomes in any given study occurs within the first 24 to 48 hours. For this reason, measurement of early mortality is relevant for prognostic purposes.

Table 3

Civilian versus Military Penetrating Injuries

Implications for Prognostic Analysis

PBI occurs in both military and civilian situations. However, major differences exist in mechanism of injury and outcome distribution in these two settings, requiring that they be analyzed separately. In the civilian setting, PBI is commonly caused by gunshot wounds; in the military setting, PBI is commonly caused by low-velocity shrapnel.

Full outcome distribution according to the GOS has only been described in two series of penetrating brain injures sustained during military action. Aarabi ²⁵ analyzed the outcome in 435 patients injured during the Iran-Iraq War and both Brandvold et al. ²⁶ and Levi et al. ²⁷ report separately on essentially the same patient population with injuries sustained in ongoing military hostilities in Lebanon. The outcome distribution of these two patient series is shown in Figure 3. When comparing this figure to the composite outcome distribution in penetrating brain injuries in the civilian situation (Fig. 2A), it is evident that the outcome in injuries sustained during military action is significantly different from civilian penetrating brain injuries.

Fig. 3

Military PBI series differ from civilian series in both mechanism of injury and severity. A number of issues explain the differences in reported outcomes between military and civilian PBI. In civilian injuries, gunshot wounds are the primary cause of PBI. In addition, among civilians, suicide is the most common cause of gunshot wounds to the brain, implying contact injuries, with high-energy transmission into the brain. A large proportion of deaths in civilian injuries occurs during the first few hours after injury.

In the military setting, penetrating injuries that reach medical attention are predominantly caused by shell and shrapnel injuries. Because of the extremely damaging nature of the cerebral injuries inflicted by the high velocity of bullets commonly found in war injuries, the majority of those suffering battlefield gunshot wounds to the head presumably never reach medical care. This skews the surviving military PBI population toward lower velocity shrapnel wounds. The priorities of field triage under battle conditions means that there is no control for those patients who are found to have a low probability of survival and therefore are not prioritized for rapid transport. Although evacuation times from the front line to military hospitals has been considerably improved in the past two decades, the average time to arrival at the hospital is considerably longer than in civilian situations. Brandvold et al. ²⁶ describe a mean evacuation time in the military setting of 2.3 hours. In civilian injures, the average time from injury to hospital is less than 30 to 45 minutes.

Additionally, the exigencies of war mean that the determinants of the methods and applications of field resuscitation and triage are difficult to control and compare with the civilian situation. There is also a much higher likelihood of wound contamination in the battlefield. The conditions under which early surgical treatment is rendered are generally unique and more limited with respect to treatment facilities in the civilian setting. For instance, although computed tomographic (CT) scanning was generally available in civilian emergency centers in the 1970s, the first reports describing routine use of CT scanning in the military situation date from conflicts in the early 1980s. Also, injuries in the military setting occur primarily in young men in otherwise excellent physical condition; this contrasts with the civilian setting, where a broader age range is represented.

Death rates are similarly much lower for patients who reach care for penetrating injuries occurring in the military setting than in civilian penetrating brain injuries. The reported case fatality rate from World War II is 10% to 13%. ^28,29 Meirowski observed a 10% mortality reporting among 673 penetrating brain wounds from Korea. On composite analysis of 2,187 consecutive penetrating wounds of the brain from Vietnam, Hammon ³⁰ describes a hospital mortality of 11%. This excludes, however, patients with injuries of extreme severity in whom the management policy was 'unoperated expectant. When these patients were included, an overall mortality of approximately 30% was demonstrated. This rate is very similar to the 26% mortality described in the series reporting on injuries sustained during the Lebanese conflict. The death rate reported by Aarabi for the Iran-Iraq War is considerably lower at 16.3%. However, this series concerns a selected patient population referred to a tertiary center for definitive treatment. The mean time from injury to arrival at this center was 49 hours (range, 7-450 hours). TABLE

Table. Evidentiary T...

Table. Evidentiary T...