Arguments have been made that these very early decisions to limit treatment at the time of initial emergency assessment are ethically appropriate (so as to avoid prolonging suffering by delivering medical care which is futile) and financially important (so as to avoid high cost medical care which has no chance to improve outcome) [3]. However, all of these considerations are predicated on the assumption that prognostication is sufficiently accurate and reliable to enable decision-making this early after an acute neurological catastrophe. This raises important concerns about how we prognosticate, how we use this information in individual patient decision-making in the emergency setting, and how we communicate this information to patients and their families. Finally, it leads to the fundamental question: Is nihilism an effective treatment strategy in neuro-emergencies?

Non-traumatic intracerebral hemorrhage remains without a treatment of proven benefit. Predictors of short-term mortality, and to a lesser degree long-term functional outcome, are relatively well described. Most ICH prediction models have found that clinical status, such as measured by the Glasgow Coma Scale (GCS) score or National Institutes of Health Stroke Scale (NIHSS), and hematoma volume are strong predictors of 30-day mortality risk and longer-term functional outcome. Other clinical predictors present in various models include age, presence and volume of intraventricular hemorrhage (IVH), infratentorial hemorrhage location, admission blood pressure, and coagulopathy [12–16]. The most commonly used ICH prediction model, the ICH Score, involves a sum score of points assigned for GCS [3–4 = 2, 5–12 = 1, 13–15 = 0], hematoma volume [≥30 ml = 1, < 30 ml = 0], presence of IVH [yes = 1, no = 0], infratentorial origin [yes = 1, no = 0], and patient age ≥ 80 [yes = 1, no =0] [14]. ICH Scores may range from 0 to 6, and each increase in the ICH Score is associated with increased risk of 30-day mortality. While the ICH Score was developed to help standardize communication and risk stratification for ICH clinical care and clinical research, we have found clinicians increasingly tempted to use this as an early triage tool. Specifically, the first author of this manuscript has had other physicians suggest that patients with an ICH Score of 4 (predicted 30-day mortality of 97% in the original cohort) should not receive critical care or inter-facility transport because of perceived futility.

There are at least two problems with this approach. First, it assumes that a 3% chance of survival constitutes medical futility. To date, the only widely accepted definitions of futility are those that include only circumstances in which treatment will not accomplish the intended goals [17]. Second, there is considerable uncertainty around point estimates from such mortality prediction models. The fact that the 95% confidence interval of the mortality estimate in the above example extends from 81% to 100% (unpublished data), emphasizes this point.

In patients with extensive traumatic injury to the brain, predictors of death or disability include low GCS score after initial resuscitation, findings of intracranial hemorrhage or swelling on CT scan, older age, abnormal pupillary function, and hypotension early after injury [18]. In general, the motor aspect is the most reliable and informative part of the GCS score. However, current TBI guidelines emphasize that a low GCS score early after injury lacks precision for precise prediction of a poor outcome. Thus, the recognition of uncertainty remains. Interestingly, there have been attempts to develop prediction models that would drive early decisions to limit care in TBI patients with a perceived poor prognosis. A mathematical model derived on 672 patients treated at a single center from 1978–1993 suggested that long-term prognosis could be sufficiently predicted at 24 hours after TBI accurately enough to terminate life sustaining treatments in patients unlikely to survive a severe head injury (GCS ≤ 8) [2]. Notably however, the overall mortality in this cohort at 6 months was 58.8% which is nearly double that of most other series of patients with severe TBI [19–23]. Whether the extremely high mortality rate in this modeling study was due to physician bias in the care of severely ill TBI patients or other factors is unclear. However, it does clearly demonstrate the importance of understanding the context in which a particular prediction model is developed and deciding whether they are likely to apply to a specific patient (or population) in which care decisions are being made.

There have been many attempts to predict outcome in comatose survivors of cardiac arrest with HIE. Numerous studies have focused on clinical, neuroimaging, laboratory, and electrophysiological predictors. A commonly cited study published in 1985 described the outcome of patients with various clinical examination findings at different time points after resuscitation from cardiac arrest [24]. Generally findings at 3 days post-arrest have been considered the most informative. Other studies have examined the likelihood of an unfavorable outcome based on a range of predictors [25]. Importantly, recent practice parameters from the American Academy of Neurology suggested that, in the absence of brain death, clinical examination findings at day 3 of absent pupil or corneal reflexes or a motor response which was absent or no better than extensor had a sufficiently low false positive rate to reliably predict extremely poor long-term functional outcome [26]. This emphasizes that even in the setting of deep coma, some period of waiting is usually desirable to clarify the persistence and validity of clinical examination findings. Whether a trial of aggressive therapy (such as moderate hypothermia [27,28]) alters these predictive parameters in hypoxic-ischemic coma is not clearly known.

Take an example of a hypothetical cohort of 100 patients with severe stroke in which 70 of them die. Now assume that the death of about 70% of these patients was preceded by withdrawal of support [32]. If some proportion (for example: one quarter) of these patients in whom support was withdrawn might actually have survived, then the “true” mortality rate of the cohort was not 70%, but rather 58%. This would mean that 12 of the 49 patients who underwent withdrawal of support died as a result of a self-fulfilling prophecy. However, it is probably not possible to determine who among the group were those specific twelve patients. Furthermore, any outcome prediction model based on such a cohort would also be based in part on the self-fulfilling prophecy that had occurred.

Why might there be uncertainty? One central tenet is that outcome prediction models (either formal or informal) are made from studies of populations of patients, but decisions to limit treatment based on poor prognosis are made on individual patients. Prognosticating outcome in an individual patient using a model developed from a group of patients is inherently uncertain. In fact, prognostic models describe a probability of a specific outcome, such as dead or alive, but an individual patient can only have one of these outcomes. It becomes obvious that if a clinical decision rule is made such as to withdraw medical support in all patients with > 90% risk of death, then now 100% die. Prognostication, or at least the application of prognostic data, has changed prognosis.

Empirical evidence suggests this theoretical concern also is a real problem in the care of patients with neurological emergencies. In a single-center study of 87 ICH patients, Becker and colleagues found that the single most important prognostic variable in determining outcome was the level of medical support provided. In fact, withdrawal of support negated the predictive value of all other variables studied. Furthermore, they found wide heterogeneity across different physicians regarding their expectation of prognosis in the same patients. They suggested that treatment limitations, especially withdrawal of life support, might lead to self-fulfilling prophecies of poor outcome [34].

It is well recognized that heterogeneity exists in the use of various aggressive treatments for ICH, such as surgical hematoma evacuation [36], and this is not surprising given the lack of a proven effective treatment and the limited number of large clinical trials which have been performed in ICH [37–39]. However, this raises the question as to whether there is also heterogeneity in the use of measures to limit care early after ICH and whether this influences outcome. The 1983 U.S. President’s Commission on Deciding to Forgo Life-Sustaining Treatment emphasized that a do-not-resuscitate (DNR) policy should ensure that the DNR order has no implications for any other treatment decisions [40]. However, in practice DNR orders are often the first step in a continuum of care limitation [41]. Additionally, variability has been found in the use of DNR orders [42,43].

Hemphill and colleagues hypothesized that the rate at which a hospital uses DNR orders within the first 24 hours of ICH admission influences patient outcome irrespective of other hospital and patient characteristics. From a California-wide hospital discharge database, 8233 ICH patients treated at 234 different hospitals were identified [44]. Early DNR orders were one of the most common interventions, with 25% of patients having DNR orders within 24 hours of hospital admission. This was much higher than the proportion of patients who underwent aggressive interventions such as surgical hematoma evacuation or ventriculostomy placement. Of note, the rate at which a hospital used DNR orders within 24 hours of ICH patient admission increased the odds of individual patient death, even after adjusting for patient characteristics such as age, use of mechanical ventilation (a surrogate for coma), and hospital characteristics such as number of ICH patients treated or designation as a teaching hospital or trauma center. Even more importantly, there was an interaction between an individual patient’s DNR status and the hospital DNR rate. This means that not only does the individual patient’s DNR status matter, but it matters which hospital the patient is DNR in. Different hospitals (and presumably physicians) used DNR orders differently and this influenced a patient’s risk of dying. These findings of early care limitations influencing outcome in acute ICH have been confirmed by others in a separate cohort of patients in Texas [45].

This type of analysis suggests several things. First, use of measures to limit treatment early is extremely common, at least in ICH. Second, it is not the DNR orders themselves that are leading to patient death. In fact, DNR orders should have no effect on patient outcome unless the patient has a cardiac arrest. Rather, high use of early DNR orders at a hospital is a marker of an overall non-aggressive approach to ICH patients in general and this suggests that there is something about the milieu of care in a hospital that influences outcome, potentially in a very negative way. Third, it clearly demonstrates that nihilism is an ineffective treatment strategy.

Likewise, for traumatic brain injury, the ability to precisely prognosticate accurately early has been questioned. Kaufmann performed a study in which 100 consecutive patients with severe TBI were evaluated to determine whether expected prognosis on day 1 was accurate [46]. Interestingly, it was found that an experienced neurosurgeon underestimated favorable one-year outcomes and overestimated poor outcomes. Notably, an experienced neuroradiologist did the opposite, overestimating favorable outcomes while underestimating poor outcomes. The authors concluded that in severe head injury, it was not possible to reliably predict outcome on the first day with sufficient accuracy to guide management at least for purposes of unilaterally limiting treatment.

It is noteworthy that these issues are not limited to acute neurological emergencies, but apply to general critical care as well. Rocker and colleagues found that physician estimates of a patient having a less than 10% likelihood of surviving to intensive care unit (ICU) discharge were associated with subsequent life support limitation. Furthermore, these estimates were more predictive of ICU mortality than illness severity itself [47]. In a different study, Frick found that physicians tended to be overly pessimistic about survival and quality of life of ICU patients. Additionally, ICU nurses tended to suggest treatment withdrawal more often than physicians for patients who ultimately survived [48].

The most important role for prognostication in acute neuro-emergencies is in communicating risk. Prognostication is intrinsically linked to the process of communication in the care of severely ill patients. It is important to contrast this emphasis on communication and shared decision-making with some of the potentially darker aspects of prognostication, especially those which might reinforce early nihilism. The use of prognostic information to limit initial and early care because of physician nihilism due to anticipated futility is potentially problematic if this is not congruent with the patient’s (or family’s) wishes to attempt aggressive care, especially if true prognosis is perhaps less certain than that assumed by the physician. Since physicians generally cannot be compelled to provide specific medical or surgical interventions which are futile, the accuracy of prognosis is central to conflicts that arise between patient (or family) wishes and physicians about the intensity of care to be attempted. An additional argument frequently made is that high health care costs might be limited by limiting intensive care at the end-of-life. While this might seem intuitively correct when considering an individual patient’s health care costs, this may not be the case. Luce suggested that the fixed costs of ICU beds, hospital wards, and personnel such as nursing and respiratory therapists outweigh the variable costs of an individual patient’s hospitalization and that the only way to truly reduce costs in this manner may be to close beds and fire personnel [53].

We strongly favor using prognostic information to support shared decision-making with patients and families early in the setting of acute neuro-emergencies, rather than as a way to usurp their autonomy and deny or force care. This usually includes explaining, and even embracing, the uncertainty inherent in prognostication. Thus, our focus is increasingly on principles of communication and setting the stage for decision-making to come.