There is insufficient evidence to support a strong recommendation about the diagnostic accuracy of instruments and models in predicting survival in pTBI.

• The SPIN score accurately informs the probability of survival in populations of patients with pTBI.

• A SPIN score incorporating INR predicts survival more accurately than one computed without coagulopathy data.
• The Baylor score has utility in discriminating patients who will survive pTBI versus those who will not.
• The Marshall score can provide excellent discriminative ability for survival of pTBI patients.

• The ABC score, Maritzburg score, Helsinki score, Stockholm score and GCS plus ISS all have outstanding discriminative ability for survival following pTBI. The WHIS provides excellent to outstanding discriminative ability in predicting survival in military populations with ballistic pTBI and in civilian populations with nonballistic pTBI.
• The Rotterdam score can provide acceptable to outstanding discriminative ability to predict survival following pTBI.
• The St. Louis Scale has a moderate PPV and a high NPV to detect survival following pTBI.
• The REMS and mREMS provide poor to no discriminative ability to predict survival following pTBI.

In the absence of direct scientific evidence, EXPERT CONSENSUS concluded that:

• aggressive care and the avoidance of nihilism are central to achieving the best possible outcomes in pTBI patients.
• Current predictive models for pTBI should NOT be used to guide initial treatment decisions in pTBI patients and should not supersede expert opinion on prognosis because they are based on retrospective analysis of patients and have not been tested in a prospective format.

No evidence or expert opinion supported distinct recommendations based on patient gender.

The St. Louis Scale for Pediatric Gunshot Wounds to the Head specifically investigated patients under the age of 19.

Only the WHIS was specifically evaluated in military and civilian populations and stratified by wounding mechanism.

Penetrating TBI has one of the highest case fatality rates of all types of injury. Mortality rates in excess of 90% for patients with gunshot wounds to the head have been reported in some series (2053059). However, true natural history data for pTBI is lacking. Given the issues of therapeutic nihilism in patients with pTBI, it is not surprising that reported rates are so high as the perception of a poor prognosis can become a self-fulfilling prophecy. Tools that allow health care teams to accurately predict mortality, or survivability, are badly needed. The ability to accurately predict who may survive their injury allows for better allocation of resources and ability to communicate accurately with families. Over approximately the past ten years, an increased body of literature has developed and validated a variety of instruments and models that might help predict survivability following pTBI. The various models use a combination of patient characteristics, injury characteristics, physical exam, radiographic findings, and laboratory values to derive prognostic models for survival following pTBI.

Previous versions of the guidelines did not investigate instruments and models to predict morbidity and mortality following pTBI. However, an extensive section of the first edition of these guidelines, published in 2001, compiled evidence supporting a wide variety of variables as prognosticators in pTBI (11505200). Age, intentionality (i.e. self-inflicted versus assault/homicide), anatomic wounding trajectory (i.e. perforating versus tangential tracts), hypotension, coagulopathy, respiratory distress, low GCS score, presence of fixed dilated pupils, elevated intracranial pressure, and anatomical features of injury on CT were all found to correlate with mortality and/or poor functional outcome. Many of these features are utilized in the instruments and models described above as variables used to calculate the predictive scoring.

Fourteen studies (15 publications, N=1,988 with penetrating injuries) assessed the diagnostic accuracy of instruments and models for predicting survival or mortality in penetrating traumatic brain injury (pTBI).1-15 Four studies evaluated the Surviving Penetrating Injury to the Brain (SPIN) score;2,4,8,14 three studies (4 publications) evaluated the Baylor cranial gunshot wound prognosis score;8,9,14,15 two studies assessed the ABC score;3,12 and one study each evaluated the Rapid Emergency Medicine Score (REMS)/modified REMS (mREMS);1 the War Head Injury Score (WHIS);7 the St. Louis Scale for Pediatric Gunshot Wounds to the Head;5 the Maritzburg Score,11 a novel model using age, Glasgow Coma Scale (GCS) score, non-reactive pupils, and posterior fossa or bihemispheric trajectory,6 and another novel model based on motor GCS and Injury Severity Score (ISS) score.14 Additionally, three studies assessed four models that limited criteria to computed tomography (CT) findings.8,10,13

As is generally accepted, an AUC of 0.5 suggests no discrimination 0.7 to 0.8 is considered acceptable, 0.8 to 0.9 is considered excellent, and more than 0.9 is considered outstanding (PMID: 20736804). All studies included civilian subjects, while one study also included a separately analyzed military group.7 Mean age ranged from 14 to 41 years, and the proportion of males ranged from 80% to 96%. Seven studies were conducted in the United States, three studies (2 populations) were conducted in South Africa, with one study each conducted in Croatia, Brazil, Finland/Sweden, and Columbia. Baseline mean or median GCS ranged from 3 to 12 in four studies reporting summary baseline GCS measures.2-5 The instrument with the largest amount of data was the SPIN score (4 non-randomized studies, N=858), which was developed using a combination of GCS, pupil response, gender, suicidal intent, whether or not the patient was transferred from another hospital, ISS, and Internalized Normalized Ratio (INR).2,4,8,14 Total SPIN scores can range from 4 to 52, with higher scores indicating a greater likelihood of survival. The initial study using the SPIN score (N=413) yielded an area under the receiver operating characteristic curve (AUROC) of 0.962 in discriminating between individuals with pTBI who are likely to survive versus those who are not.4 Most individuals (98%) with a SPIN score of 35 or greater survived. Validation studies of the SPIN score (cutoff: not reported) indicated similar discriminative ability; one study (N=87) reported an AUROC of 0.960,8 one study (N=257) reported an AUROC of 0.880,2 and an AUROC of 0.962 was reported in another study (N=101).14 One study (N=362) reported an AUROC of 0.820 using the SPIN score omitting the INR,.2 while a second study (N=101) reported an AUROC of 0.948 for the same measure.14 Discrimination abilities were similar when survival to 6 months was examined. These studies provide moderate strength evidence that the SPIN score provides excellent to outstanding (based on AUC Values) ability to discriminate between individuals with pTBI who are likely to survive and those who are not. Two studies provide low strength evidence that the SPIN score minus INR can discriminate slightly less well than the full SPIN score.

The Baylor score (range 1 to 5 with higher scores indicate greater likelihood of mortality) is based on GCS, pupil response, age, and bullet trajectory. One study (N=297) reported discriminative ability of 0.8756,9 and another study (N=87) reported an AUROC of 0.895. These studies provide low strength evidence that the Baylor score has excellent discriminative ability to determine pTBI patients likely to survive and those who are not.

The ABC score, based on admission blood pressure (A), brain spillage (B), and presence or absence of consciousness (C), yielded an outstanding discriminative ability (AUROC 0.968), in one study reporting AUROC (N=102).3 Another study (N=214) did not provide diagnostic accuracy information but reported the proportion of patients who died having each ABC score, from 4% mortality with an ABC score of 0 to 78% mortality with an ABC score of 5.12 The first study provides very low strength evidence that the ABC score has outstanding discriminative ability.

The REMS is based on GCS, age, heart rate, respiratory rate, and mean arterial pressure (MAP) (range 0 to 26 with higher scores indicating greater likelihood of mortality, cutoff not reported) performed much less well in one study (N=134), with AUROC 0.583 indicating little discriminative ability to determine likelihood of mortality versus survival.1 The modified REMS (mREMS), which uses systolic blood pressure instead of MAP and includes mechanism of injury, demonstrated no discriminative ability (AUROC 0.506).1 This study provides very low strength evidence that the REMS and mREMS provide poor to no discriminative ability.

The WHIS is based on the GCS and ISS score, and was used in one study that reported results for a military population with ballistic pTBI (N=43) and a civilian population with non-ballistic pTBI (N=41).7 Using a cutoff score of 10, the WHIS yielded an excellent discriminative ability with an AUROC of 0.88, a sensitivity of 88%, and a specificity of 76% in the civilian sample, and an outstanding AUROC of 0.92, a sensitivity of 100%, and a specificity of 78.8% in the military sample. This study provides very low strength evidence that the WHIS provides excellent to outstanding discriminative ability based on population.7

The Maritzburg score, based on the motor GCS and brain spillage, was discussed in the same population as the ABC score above that reported AUROC (N=102), and had outstanding discriminative ability to determine those likely to survive versus those likely not to survive (AUROC 0.94).11 This study provides very low strength evidence of outstanding discriminative ability for mortality/survivability of the Maritzburg score.

The Rotterdam score, the Marshall score, the Stockholm score, and the Helsinki score rely solely on findings from head CT. The Marshall score is based on the degree of cerebral swelling (i.e., midline shift, compression of basal cisterns) and the presence/size of contusions/ hemorrhages. The Rotterdam score also includes midline shift and the degree of basal cistern compression but does not include the presence of contusion and limits other criteria to the presence of an epidural mass lesion and intraventricular blood/traumatic subarachnoid hemorrhage (SAH). The Stockholm score is based on SAH in convexities, basal cisterns, intraventricular hemorrhage, midline shift, epidural hemorrhage, diffuse axonal injury, and dual-sided subdural hematoma. The Helsinki score is based on the presence of subdural hematoma, intracerebral hematoma, epidural hematoma, hematoma volume, intraventricular hemorrhage, and compression of basal cisterns. One small study (N=87) reported the discriminative ability of the Marshall and the Rotterdam scores (Marshal: AUROC 0.874; Rotterdam: AUROC 0.905) for determining the likelihood of survival or not.8 Another small study (N=75) reported the ability of all four CT-based instruments to discriminate between 6-month mortality versus survivability: Marshall AUROC 0.815; Rotterdam AUROC 0.774; Stockholm AUROC 0.850; Helsinki AUROC 0.900.13 The Helsinki score (reference score) outperformed the Marshall score (p=0.046) and the Rotterdam score (p=0.003), but not the Stockholm score (p=0.089). Among treated patients only (N=59), the discriminative ability dropped somewhat: Marshall AUROC 0.750; Rotterdam AUROC 0.654; Stockholm AUROC 0.783; Helsinki AUROC 0.816.13 Again, using the Helsinki score as the reference, the Helsinki score performed better than the Rotterdam score (p=0.037), but not the Marshall score (p=0.362) or the Stockholm score (p=0.390). All scores (N=71) had similar discriminative ability for unfavorable outcome, defined as Glasgow Outcome Scale score 1 to 3 AUROCs 0.800 to 0.887.13 One study (N=95, includes 9 tangential TBI and 19 patients with unknown TBI status) did not provide diagnostic accuracy data but did report mortality by Marshall score: I-III 20.2%, IV: 15.2%; V-VI 1.3%.10 These studies provide: low strength evidence that the Marshall score can provide excellent discriminative ability; very low strength evidence that the Rotterdam score can provide acceptable to outstanding discriminative ability; very low strength evidence that the Stockholm score can provide excellent discriminative ability; and very low strength evidence that the Helsinki score can provide outstanding discriminative ability.

The St. Louis scale for pediatric gunshot wounds to the head is based on pupil response, bullet trajectory, and lobes of injury. One study (N=71) did not report discrimination ability of the St Louis scale, but reported that with a cutoff score of 5 or above the scale had a sensitivity of 94.12%, specificity of 75.68%, positive predictive value (PPV) of 78.05%, and a negative predictive value (NPV) of 93.33%.5 This study provides very low strength evidence that the St. Louis Scale has a moderate PPV and a high NPV.

Another study (N=101) reported outstanding discriminative ability using the continuous motor GCS plus ISS score (AUROC 0.951).14 When assessing the model's ability to discriminate between survival and mortality at 6 months, the same study (N=35) reported an AUROC of 0.937. This study provides very low strength evidence that GCS plus ISS can provide outstanding discriminative ability.

A novel scoring system developed by Gressot et al., (N=119) using age (>35 years), GCS score (3 or 4), nonreactive pupils, and posterior fossa or bihemispheric trajectory was found to correlate with mortality outcomes; while measures of diagnostic accuracy were not reported, patients with scores of 3 to 5 had a 75% mortality rate and patients with scores of 0 to 1 had a 25% mortality rate (50% mortality in patients with score of 2).6 As measures of diagnostic accuracy were not reported in this study, we did not rate the strength of evidence.

Although many of the instruments described above have good accuracy for determination of survivability following pTBI, all are imperfect, especially when predictions for a single patient are considered. The SPIN score is currently the most studied and validated model although numerous models performed well in initial investigations.

When using these prediction models, their inherent limitations must be considered. Predictive modeling is based on probabilities across a population and therefore carry inherent inaccuracies when applied to individual patients especially in the acute post-injury time-frame. Additionally, the type of data used to generate these models must be considered. As most modern pTBI patients die from withdrawal of life-sustaining therapy, prognostic models may not be based on natural history data but instead a perception of a bad outcome. It is also important to consider the relationship between prediction models and nihilism. The SPIN score was developed with an interest in reducing nihilism, demonstrating that patients with pTBI can achieve acceptable outcomes with aggressive care. By contrast, however, the use of prognostic calculators has been associated with nihilism (19218017, 29239321).

It should also be noted that this key question focused only on survival and not on functional outcome which is clearly a subject of critical importance to patients and their families. It is critical to frame family meetings with discussions of functional outcomes which are as important, if not more important, than just survival in determining the extent of the care rendered. Care should be taken in discussion with families when issues of survival or functional outcome are considered. As a general rule, the terms survivability - which indicates likelihood of whether the patient will live or die - and salvageability - which is related to a likelihood of a poor functional outcome - may be considered.

The use of prognostic calculators generated a substantial amount of discussion amongst our panelists and, in particular, there was great concern that the predictions they generate could be used or misused to limit care of patients who might have a chance of an acceptable outcome. Our panelists expressed concern about nihilism across the breadth of neurotrauma but felt that problems with nihilism were more significant as it relates to pTBI. Concern was expressed that many clinicians currently judge care for those with a gunshot wound to the head as automatically futile despite the fact that many of these patients can have good neurological outcomes. Indeed, in considering the fact that military patients have better outcomes than civilian patients despite higher caliber weapons in combat, it is believed that the highly aggressive approach taken to pTBI in the military is likely the key factor underlying this difference.

Admittedly, the high lethality of suicide-related pTBI is likely also a factor. Although there is very little published data regarding nihilism, our group felt a strong need to combat inappropriate therapeutic nihilism despite our limited ability to write evidence-based guidelines on the topic. Acknowledging that the TQIP guidelines recommend at least 72h of aggressive care before making withdrawal of life sustaining therapy decisions, our group felt that this timepoint may be premature given that important neurological improvements are often seen after this time point.

Our group ultimately felt that existing prognostic calculators should not be used to drive care-limiting decisions. While the information inherent to prognostic calculations can provide a helpful benchmark and can have utility is educating substitute decision-makers, the imprecision of these tools in the care of an individual patient must be considered.

Prognostication has been a major recent advance in TBI care. There will be a perpetual push to develop increasingly accurate yet simpler models that are accurate earlier and earlier in the course of care. As models become more accurate it may be reasonable for them to play a greater role in patient care decisions. Machine learning algorithms, for example, are starting to out-perform regression models and are anticipated to advance pTBI prognostication in the near future.

The enthusiasm to develop prediction models has not been equaled by efforts to delineate how to use prognostic information. What is a good outcome? What is an acceptable chance of achieving a good outcome? How accurate does a prognostic calculator need to be to support routine clinical use? A tremendous amount of additional research is needed in this regard. Implicit in the term nihilism is the assertion that physicians play a key role in medical decision-making in critically ill patients. The extent to which decision-making in these patients is made by physicians vs. well-informed substitute decision-makers is also very deserving of discussion. It would also be beneficial for prognostic models to incorporate longer-term functional outcomes as opposed to only short-term survival.