Predicting Mortality and Outcomes after Traumatic Brain Injury with Prognostic Models

The Corticosteroid Randomization After Significant Head Injury (CRASH) trial and International Mission on Prognosis and Analysis of Clinical Trials (IMPACT) models were some of the first multifactor models used in the prediction of TBI outcomes. These models aimed to give clinicians tangible, clinically based, accessible data to predict mortality when patients arrived at the hospital with a TBI. The CRASH models were derived from the database of 10,008 patients with Glasgow Coma Scale (GCS) less than 14 who had been enrolled within 8 hours of injury for the Medical Research Council trial of corticosteroids after significant head injury (MRC CRASH) clinical trial, recruited between 1999 and 2004. The aim of the CRASH model was to predict death within 14 days and death and disability within 6 months for these patients. Two models were developed, 1 using only patient clinical characteristics on admission and another with the addition of computed tomography (CT) scan findings. The CRASH and CRASH-CT models were externally validated using the IMPACT dataset. The validated models used clinical predictors such as age, sex, GCS score, CT scan findings, and concomitant injuries and found that factors such as age greater than 40, lower GCS, absent pupillary reactivity, and obliteration of the basilar cisterns or unevacuated hematoma portend a worse prognosis ( Table 1 ). The models also found that patients in low-income to middle-income countries had worse prognosis than those in high-income countries. The models showed great discriminatory ability overall based on a C statistical of 0.8 and a C statistical of 0.77 when externally validated with the IMPACT dataset. The limitations of these models include the confounders present in low-middle income versus high-income countries, such as low-middle income countries having fewer resources and younger patients who presented to the hospital later after injury. Additionally, the IMPACT dataset did not include low-middle income countries, limiting use of the IMPACT dataset for external validation.

The IMPACT dataset subsequently was used to create a prognostication model that used a larger set of clinically available admission characteristics. The IMPACT dataset included 8509 patients greater than 14 years old with GCS less than 12 from 11 studies conducted between 1984 and 1997. ^, The IMPACT models used similar prediction factors to the CRASH models, but with 3 progressive iterations: core, extended, and laboratory models. The core IMPACT model used age, motor score, and pupillary reactivity for the predicted 6-month Glasgow Outcome Scale Extended (GOSE), and the extended model included secondary insults such as hypoxia, hypotension, and CT characteristics and pathologies. The laboratory model added laboratory values of glucose and hemoglobin to the features of the extended model. The IMPACT models showed that age, GCS motor score, and pupillary reactivity were the greatest predictors of poor outcomes, with other relevant predictors including GCS motor scores of 1 to 2, CT scans demonstrating traumatic subarachnoid hemorrhage or raised intracranial pressure, and abnormal laboratory values such as hypoglycemia and anemia. As each model increased in complexity, so did the discriminatory values, with observational studies having a maximum area under the curve (AUC) greater than 0.8 when cross validated. While using the patients enrolled in the MRC CRASH trial for external validity, they found the AUC was 0.776 for mortality and 0.780 for unfavorable outcome.

Despite the limitations in overall generalizability of the final CRASH and IMPACT models, these early neuroprognostic tools were available as web-based calculators to aid in clinical decision-making, set a high-standard for creating validated neuroprognostic models, and remain the only readily available neuroprognostic tools. Recently, data from the contemporary Transforming Research and Clinical Knowledge in TBI (TRACK-TBI) observational clinical trial was used to validate the IMPACT and CRASH models. It was found that the IMPACT and CRASH models overpredicted mortality and CRASH overpredicted unfavorable outcomes, but had good discriminatory abilities. In addition to the predictive factors discussed above, other studies have found that adding International Normalized Ratio (INR), total leukocyte count, and need for blood transfusion resulted in a slight improvement in the ability to predict 6-month outcome in TBI. These contemporary evaluations highlight the need for updated prognostic models that include a wider variety of clinical variables, a broader understanding of recovery, and use advances in biomarkers and neurocritical care to inform decision-making.

Defining Disability and Recovery After Traumatic Brain Injury

The aforementioned studies, and many others exploring recovery after TBI, define disability based on the GOSE. The GOSE defines outcomes as death, vegetative state, lower or upper severe disability, lower or upper moderate disability, and lower or upper good recovery based on the ability to complete normal tasks of daily living, socialization, and incorporation into life after injury. However, the ability to take each of these standardized categories, eloquently apply them to individuals with TBI, and discuss them in informed ways with surrogate decision makers is a challenge.

The tangible recovery milestones incorporated in the GOSE include recovery of consciousness, ability to follow commands, and ability to independently perform activities of daily living. Speaking with families regarding recovery of consciousness requires a definition of consciousness and coma. The Second Curing Coma Campaign defined consciousness as “…demonstrat(ing) the global neuronal integration with the ability to rapidly swap states and occupy functional states or brain configurations outside the constraints of anatomic connectivity.” Biologically, consciousness requires functioning of the rostral brainstem, midline thalamic nuclei, anterior insula, and cortical interconnections, which can be altered by injury and sedating medications. When recovering consciousness, it is imperative to allow time for regaining these networks.

Studies have shown that persistent disorder of consciousness (DOC) is associated with a high velocity injury, younger age, mass effect greater than 5 mm, intraparenchymal contusions, interventricular hemorrhage (IVH), and longer acute care stay, whereas the recovery of consciousness is associated with the absence of IVH, male sex, less severe white matter injury or mass effect, and shorter acute stay. Time to follow commands was increased by 9.2 days with presence of diffuse axonal injury (DAI) and increased 6.4 days with the presence of IVH. Acknowledgment and awareness of these clinical variables can aid in discussions with family about expected time to regain consciousness.

Although data for severe TBI suggests a mortality of 12% to 44%, almost 36% is attributed to early withdrawal of life-supporting therapies. For example, in a TRACK-TBI study on moderate and severe TBI, 33% of mortalities occur within 72 hours, 52% within the first week, and 70% within the first 2 weeks. In patients who survived or for whom withdrawal of care was not pursued, a recent study of moderate to severe TBI patients aged 18 to 79 years with a GCS motor score less than 6, demonstrated that 12% continued to have a disorder of consciousness by rehabilitation; however, 82% of those patients recovered consciousness by the end of rehabilitation. By completion of rehabilitation, 2% of the initial cohort had persistent DOC (GCS motor scores below 5). Of note, 4% of the initial cohort who presented following commands regressed to a DOC on admission to rehabilitation. In terms of functional recovery, 80% of patients without DOC at discharge from acute stay and 40% of patients with DOC regained semifunctional or fully functional independence. Discussion of a timeline that anticipates recovery of consciousness may not occur until after the initial inpatient hospital stay offers the potential to streamline decision-making for families.

It can be helpful for families to be given predictions of recovery beyond the acute inpatient hospital or rehabilitation period. For a severe TBI, data show that 2 weeks post injury, one can expect 12.4% to have a GOSE score of 4 to 8, but only 0.7% with near full functioning (GOSE 7–8); for moderate TBI, 41% had a GOSE that was 4 or better. At 6 months, the proportion of patients with a GOSE 4 or greater increased to 45% for severe TBI and 71% for moderate TBI—effectively quadrupling and doubling, respectively. Recovery seems to slow after 6 months, with 52.4% severe TBI and 75% moderate TBI patients ultimately attaining a GOSE 4 or greater at 1 year. These statistics are important to communicate to families, as those patients given sustained treatments have the opportunity to recover if given time and appropriate rehabilitation.

Recovery beyond 1 year and reintegration into society can be difficult to predict due to limited data. In a study by Deng and colleagues, the median time to follow commands in severe TBI patients aged 16 to 80 was 8.5 days after injury. At 3 months post injury, the mean GOSE was 4; GOSE was 5 at 6 months to a year, and 6 at 2 years. At 2 years post injury, 64.3% of patients had a GOSE of 4% to 6% and 50% of patients had a GOSE 7 to 8, which represents recovery to near-baseline abilities. Such data can provide families with some understanding of potential recovery beyond 1 year.

Using Data to Aid in Decision-Making

The decision to continue with care or pursue withdrawal of life-supporting therapies typically occurs when surrogate decision-makers are approached to consent to long-term ventilation and nutrition procedures. At this time, discussion can be aided by prognostic and decision tools. It is not uncommon in patients with moderate to severe TBI to have a tracheostomy. In fact, a study showed that 38.1% of patients had a tracheostomy, with most patients with prolonged ventilation eventually advancing to rehabilitation.

When discussing with surrogate decision-makers the need for tracheostomy and semipermanent enteral nutrition, the rationale for the procedures, and the opportunity for recovery should be discussed with them. Patients with TBI have loss of protective reflexes such as cough and gag putting them at risk for aspiration, central apnea, and sympathetic overdrive causing tachypnea. These issues—in addition to contributions attributable either to direct damage to the pons or medullary respiratory centers or weakness of ventilatory muscles due to prolonged ventilation—place them at a high risk for prolonged intubation and/or reintubation. Concomitant injuries that increase the chance for needing long-term mechanical ventilation include aspiration, concurrent pneumonia, pulmonary edema, or other traumatic injuries. ^, Increased risk of needing tracheostomy is associated with older age, increased trauma severity, surgical intervention, GCS less than 8, thoracic trauma, hypoxemia, and nonreactive pupils. ^,

There is conflicting data regarding the benefits of early versus late tracheostomy. Some advantages of early tracheostomy include reduction in respiratory dead space, decreased sedation, decreased ventilator associated pneumonias, and an earlier transition to rehabilitation, which could be associated with improved long-term outcomes. However, this must be balanced against evidence that there is no difference in overall mortality, length of stay, duration of ventilation, or pneumonia in early versus late tracheostomy. ^, Mubishir and colleagues analyzed data for hospitalized TBI patients over a 20-year period spanning 1995 to 2015 relative to tracheostomy timing and outcomes, finding that earlier tracheostomy (7 days or sooner) was associated with a shorter length of stay (27 days vs 48 days) than late tracheostomy (greater than 15 days). Overall, 5.6% of patients had tracheostomies; the procedure was associated with a 35% lower risk of in-hospital mortality versus no tracheostomy. Of note, patients who underwent early trach at 3 to 5 days post injury had higher mortality when compared to a late tracheostomy, but the underlying reason for this finding was not clear and may reflect injury severity or concomitant trauma. Conversely, a CENTER TBI study defined early tracheostomy as less than or equal to 7 days from admission and late tracheostomy as greater than 7 days and found that the timing of tracheostomy had no significant effect on ICU mortality, 6 month mortality, or GOSE; however, late tracheostomy was associated with longer ICU and in-hospital stay. It was suggested that each additional day between hospitalization and tracheostomy was associated with a 4% increase in unfavorable outcomes.

Ideally, discussions regarding goals of care should occur no earlier than 72 hours after resuscitation and stabilization, take place in multiple sessions to build trust, and consider patient values, social support and economics. ^{, ,} Current guidelines for neuroprognostication recommend a general recovery trajectory be formulated for the patient, and further discussions on tracheostomy and semipermanent nutrition can be approached.

The effectiveness and productivity of family conversations may be augmented with decision aids. Decision aids are commonly used in other specialties to help families understand options in care and to facilitate discussions with the care team. It is recommended that decision aids be printed on paper and written at a 6th grade reading level, with introduction and icons for prognosis description, 6 to 12 month validated outcomes, concomitant injuries, and delineation of discharge destinations. ^, In a study surveying families and surrogate decision-makers with experience in using decision aids for goals of care discussions, surrogate decision-makers preferred numeric estimates that provide objectivity as well as provider compassion, but noted that it was important that compassion did not give false hope. They valued acceptance of their decision and warned that prolonging treatment in the hope of more positive outcomes may only deepen grief when an unfavorable outcome occurs—emphasizing the appreciation of blunt pessimism when needed.

Interestingly, when decision aids were used, withdrawal of life-supporting therapy decisions was made only 27% of the time, versus 56% in a control group. In this specific study, at 3 months the decision regret was low in both groups, but slightly higher among those who used the decision aid.