Approximately 15-20% of mechanically ventilated patients admitted to intensive care units (ICUs) have an admission diagnosis involving a neurocritical care condition.1 Compared with other critically ill populations, those with various forms of brain injury generally have worse outcomes. Apart from having a higher risk of death, those who survive are frequently left with various degrees of functional and cognitive limitations.2-8

Only a small number of specific interventions have been definitively demonstrated in large randomized controlled trials to improve outcomes in neurocritical care.9-14 Many of the treatments that are recommended in consensus guidelines and incorporated into protocols are, by necessity, based on low-quality evidence derived largely from observational studies.15-19 For example, over the past decade it has become common practice to prevent/treat aggressively even relatively mild derangements in core body temperature, blood glucose levels, and serum sodium and hemoglobin concentrations.20-23 With increasing use of electroencephalography, it has been recognized that many neurocritical care patients develop non-convulsive seizures, which most experts believe should be prevented or expeditiously diagnosed and treated.24 Care of patients is increasingly being individualized using multi-modal monitoring to adjust physiological goals, including intracranial pressure (ICP), cerebral perfusion pressure, and PCO2 or PO2 targets.25-27 Several observational studies have suggested that an organized, protocol-based approach with emphasis on the prevention of “secondary” neurological insults may be associated with improved recovery.28-34

Consistent with these developments, the care of neurocritical care patients in our region of Canada has evolved considerably over the past decade. There are few contemporary data assessing whether outcomes have changed. We performed a cohort study in Southern Alberta to assess whether hospital mortality and discharge disposition have improved. If they had, we wanted to explore whether certain practice modifications contributed and if the improvements were consistent across various diagnostic subgroups. Information concerning the natural history and prognosis of neurocritical care patients has important implications for bedside clinicians, health care administrators, and patients’ families.

Methods

The Research Ethics Board at the University of Calgary approved the study on June 28, 2012. The need for informed consent was waived, given that it was considered a quality assurance project. Explicit permission to publish results was provided.

The Alberta Health Services (Calgary Zone, previously Calgary Health Region) Department of Critical Care Medicine is responsible for administering critical care services at all hospitals for adults in Calgary. There are four closed ICUs located in three hospitals, including three “multi-system” and one post-cardiac surgical unit, all of which are staffed at all times by fellowship-trained critical care physicians. All critically ill patients with trauma, stroke, and/or neurosurgical conditions are cared for in an ICU at a single hospital, which serves as the only trauma and neurosurgical referral center for Southern Alberta. Some cardiac arrest patients, largely with acute coronary syndromes, are cared for in a coronary care unit. They are not included in this study.

Since 2001, information from all patients admitted to regional ICUs has been collected prospectively in a departmental database. At the time of ICU admission, attending critical care physicians are responsible for recording primary and secondary admission diagnoses using the Intensive Care National Audit & Research Centre coding method. This comprehensive system of recording admission diagnoses has been validated and demonstrated to have good inter-observer reliability.35 Each patient is given only one primary admission diagnosis but can be assigned multiple secondary diagnoses. The specific diagnostic codes used are listed in the Appendix. In addition, intensivists routinely document the admission Glasgow Coma Scale (GCS) score and relevant “chronic health points”, which in turn are combined with the most abnormal physiological data from the first 24 hr to calculate the Acute Physiology, Age, and Chronic Health Evaluation II (APACHE II) score.36 Computer specialists regularly monitor data entries and, if necessary, contact the attending physician to minimize the occurrence of missing data.

Using our database, we identified consecutive neurocritical care patients admitted to local ICUs between June 13, 2001 (start of the database) and July 28, 2012. We defined neurocritical care patients as having one of the following primary or secondary admission diagnoses, as determined by the admitting critical care physician: traumatic brain injury (TBI), subarachnoid hemorrhage (SAH), intracerebral hemorrhage (ICH), ischemic stroke, anoxic encephalopathy, central nervous system (CNS) infection, or status epilepticus. Patients with anoxic encephalopathy were included because by far the most common cause of death for survivors of out-of-hospital cardiac arrest is failure to awaken owing to severe neurological injury.37

The primary outcome of interest in this study was hospital mortality. Using information from the Alberta Health Services discharge abstract database, we were also able to determine whether patients had been discharged home (with or without support services), to another institution, or to a long-term nursing care facility. This information has been recorded prospectively since April 1, 2002 by each hospital’s Health Records Department. The discharge codes are listed in the Appendix. The two data sets—the first containing clinical data and the second discharge information—were linked using the patients’ unique medical record numbers. For the purposes of this study, we defined a “favourable” neurological recovery as one where the patient was discharged home without the need for support services.

To assess the association between the date of admission and eventual outcome, we divided the entire epoch into four shorter time periods with an equal number of days (n = 1,015 each): Period 1 was June 13, 2001 to March 25, 2004; Period 2 was March 26, 2004 to January 4, 2007; Period 3 was January 5, 2007 to October 16, 2009; Period 4 was October 17, 2009 to July 28, 2012).38 The rationale for dividing the data in this fashion was to minimize the impact of natural fluctuations in outcomes that might occur during shorter time intervals. We then used multivariable logistic regression models to adjust for patient age and the initial level of consciousness (assessed using the GCS score), both of which are known to be directly associated with mortality. These potential confounders were selected a priori because they are well known to be strongly associated with outcomes among neurocritical care patients with a variety of diagnoses.39-42 We also adjusted for diagnostic case mix (TBI vs other diagnoses), compared temporal trends between different diagnostic subgroups, and in a separate analysis considered time as a continuous variable (per 365-day block rather than for longer time periods). We also compared outcomes before and after certain developments in our region: availability of fellowship-trained neurocritical care specialists for consultation (September 2003); implementation of a temperature regulation protocol with availability of advanced cooling technology (endovascular catheters) (September 2004); mutual care rounds involving the ICU, neurocritical care, and neurosurgical services (July 2005); use of a comprehensive TBI management protocol (August 2008); and clustering of patients with neurocritical care diagnoses in a neurocritical care “pod” within a larger multidisciplinary unit (June 2010).

Because the APACHE II score includes age and GCS score as covariates, we calculated a modified score from which their contribution was subtracted. The APACHE II score is calculated based on the “worst” physiological values during the initial 24 hr in the ICU. It is generally considered to be a marker for baseline severity of illness. However, because the early resuscitation and care of critically ill patients can influence the 24-hr APACHE II score, it could also be viewed as an intermediate variable. Indeed, more effective early resuscitation could conceivably lead to a lower APACHE II score and, in turn, better outcomes. Given that statistical adjustment for an intermediate variable is inappropriate, we initially created our multivariable models without including the modified APACHE II score.43 However, we also performed a separate secondary analysis wherein we adjusted for it.

Statistical analysis was performed using SAS software (version 9.3; SAS Institute, Cary, NC, USA). Categorical variables were presented as proportions and compared using chi-square analysis. Continuous variables were presented as medians with the interquartile range and were compared using the Kruskall-Wallis test. Multivariable odds ratios (ORs) were presented with 95% confidence intervals (CIs). In cases where data were missing, the values were not replaced. A P value of < 0.05 indicated statistical significance.

Results

Patient characteristics

Between June 13, 2001 and July 28, 2012, a total of 4,097 patients admitted to regional ICUs had a primary or secondary neurocritical care diagnosis at admission. Patient characteristics are shown in Table 1. There was a somewhat lower proportion of patients with TBI and a higher proportion with status epilepticus during the most recent time period than during previous periods. Age and median GCS scores varied marginally among time periods but not in a consistent direction. APACHE II scores decreased slightly over time.

Table 1 Characteristics of neurocritical care patients according to time period of admission
Full size table

Patient outcomes

Hospital mortality data were available for all patients. The proportion of patients dying in hospital decreased significantly over time, from a maximum of 40% during the first time period to a minimum of 32% during the most recent epoch (Table 2; P = 0.001 for differences between the four time periods).

Table 2 Outcomes of neurocritical care patients according to time period of admission
Full size table

Discharge disposition data were available for 3,493 patients (85% overall, 91% since discharge abstracts were collected). The proportion of patients discharged home without the need for support services increased over time, from a minimum of 33% during the first period to 39%, 38%, and 40%, respectively, during subsequent time periods (Table 2; P < 0.001 for differences between time periods).

The median ICU stay among hospital survivors increased over time, ranging from 3.8 days for the first two time periods to 4.7 and 4.6 days, respectively, for the latter two time periods (Table 2; P < 0.01 for the difference between time periods). In contrast, among patients who died, the ICU stay did not change.

Multivariable analysis

With adjustment for patient age, initial GCS score, and diagnostic case mix (non-TBI vs TBI), there was a gradual reduction in the odds of hospital death from one time period to the next (Table 3). This association persisted with inclusion of the modified APACHE II score in the model as a potential confounder, rather than considering it to be an intermediate variable (OR 0.77, 95% CI 0.60 to 0.98, P = 0.03). A similar result was obtained when the admission date was assessed as a continuous variable (OR per year 0.96, 95% 0.94 to 0.99, P < 0.01).

Table 3 Multivariable analysis assessing the association between time period of admission and outcomes (adjusted for age, GCS score, and case mix)
Full size table

As expected, patient age and the initial GCS score were powerful predictors of hospital mortality. Patients with diagnoses other than TBI had a higher mortality rate than patients with TBI.

A similar multivariable analysis, adjusting for the same covariates, was performed using “discharge home without the need for support services” as the outcome of interest (Table 3). The odds of being discharged home were significantly higher for the most recent time period than for the first time period. This was also true when the APACHE II score was included in the model (OR 1.40, 95% CI 1.09 to 1.79, P = 0.009). Using the admission date as a continuous variable, the OR (per year) was 1.05 with 95% CI 1.03 to 1.08 (P < 0.0001). When the analysis was restricted to patients who survived, the survivors during the last time period were significantly more likely to be discharged home without the need for support services (OR 1.41, 95% CI 1.08 to 1.85, P = 0.01).

A pre-specified subgroup analysis was performed to assess the change in mortality over time (last time period vs first time period) according to the admitting diagnosis. Statistically significant reductions in mortality were observed for patients with TBI (OR 0.63, 95% CI 0.42 to 0.95, P = 0.03) and SAH (OR 0.49, 95% CI 0.25 to 0.95, P = 0.03). Comparable trends were observed for patients with anoxic brain injury, ICH, and status epilepticus but not for patients with ischemic stroke or CNS infection (Figure). When we repeated the multivariable analysis and included an interaction term for diagnostic category and time period of admission, there was evidence of a differential effect of time depending on the specific diagnosis (P < 0.0001).

Figure
figure 1

Multivariable analysis assessing odds of hospital mortality during the most recent period compared with the initial time period based on the admitting diagnosis. Abbreviations: ICH = intracerebral hemorrhage; IS = ischemic stroke; SAH = subarachnoid hemorrhage; SE = status epilepticus; TBI = traumatic brain injury

Full size image

Additional multivariable analyses were performed to assess mortality and discharge disposition before and after implementation of various changes in our region (Table 4). Each change that was introduced was associated with a favourable trend in subsequent outcomes.

Table 4 Multivariable analysis assessing outcomes before (reference) and after implementation of various protocols and changes to the model of care
Full size table

Discussion

There has been a significant reduction in hospital mortality among neurocritical patients in Southern Alberta over the past 11 years. The proportion of patients discharged home without the need for support services (home care) has also increased. These changes persisted even with adjustment for variability in patient characteristics over time, including important prognostic factors such as age, initial level of consciousness (GCS score), diagnostic case mix (TBI vs other diagnoses), and severity of early physiological instability (APACHE II score). Improvements in outcomes were more obvious in some diagnostic subgroups than in others.

There are probably multiple reasons for improved outcomes in our region over time. During these 11 years, relevant changes have occurred, including hiring fellowship-trained neurointensivists to staff a consultative neurocritical care service; development of a temperature-regulation protocol, with increasing use of advanced technology (endovascular cooling catheters) aimed at achieving normothermia in most neurocritical care patients and therapeutic hypothermia in selected patients; collaborative morning rounds between ICU, neurocritical care, and neurosurgical services; adoption of a standardized regional protocol for the care of patients with severe TBI, with increasing use of multi-modal neurological monitoring (e.g., ICP, brain tissue oxygen tension, pressure reactivity, continuous electroencephalography, somatosensory evoked potentials, cerebral microdialysis) and a staged algorithm for treating intracranial hypertension; and clustering of neurocritical care patients (within a larger multi-system ICU) at the neurosurgical and trauma center (Foothills Medical Center) based on admission diagnosis. These changes in practice were temporally associated with reduced mortality and a greater proportion of patients discharged home without support (Table 4). Although we have provided specific dates in our analysis for when these changes were implemented, the principles underlying some of them were introduced more gradually. For example, many of the aspects of the local TBI management protocol had been embraced years before the protocol became an official document.

Over the past decade, there have been other gradual changes in practice that were not assessed in our analysis largely because there was no one specific time when they were implemented. Relevant examples might include increasing use of endovascular coil embolization to treat cerebral aneurysms; use of decompressive craniectomy to treat refractory intracranial hypertension and herniation syndromes; increasing availability of neurosurgical or neurological consultants with subspecialty interests; and more neurorehabilitation beds. Most of these changes in practice are not specifically supported by randomized controlled trials that have demonstrated their efficacy. Some practices, such as the use of hypothermia in TBI patients, decompressive craniectomy, and ICP monitoring, have been reported to lack efficacy at improving outcomes in preliminary clinical trials.44-46 Nevertheless, our findings are in agreement with multiple observational studies suggesting that the development of specialized neurocritical care units—usually supported by the presence of neurointensivists and utilizing evidence-based protocols—may contribute to improved patient outcomes.28

International studies involving neurocritical care patients have suggested that there is considerable variability in outcomes between countries and individual centers. Mortality rates have not consistently decreased among patients with TBI over the past two decades.47,48 Conflicting conclusions have been reached among patients with SAH and ICH, although trends have probably been favourable.49-53 There are relatively few corresponding Canadian data. A recent large national study involving more than 13,000 patients who had been resuscitated and hospitalized following cardiac arrest did not detect any improvements in survival rates over time.54 A study from Ontario using ICD-9 codes to identify hospitalized patients with TBI found no reduction in mortality between 1992 and 2001.55 Administrative databases have also been used to identify patients with hemorrhagic stroke from across Canada and to compare outcomes between 1982-1983 and 1991-1992. Statistically significant improvements in case-fatality rates were reported for patients with ICH but not SAH.56,57 In contrast to these findings, we have found substantial reductions in mortality rates for neurocritical care patients in Southern Alberta over the past decade.

Most deaths in critical care patients occur following the withdrawal of life-sustaining interventions.58,59 Opinions concerning the prognosis of neurocritical care patients vary among critical care practitioners. This variability may lead to significant differences in withdrawal-of-care practices,60-64 which in turn may explain some of the marked variability in outcomes that has been described, even within Canada.54,60 Some research has suggested that physicians may be generally more pessimistic about the prognosis of neurocritical care patients than is justified by their particular clinical characteristics.65 We found that the length of ICU stay among hospital survivors has increased over time, possibly indicating that local critical care physicians are delaying decisions related to withdrawal of life-sustaining interventions more often than previously. It is conceivable that this delay may have contributed to the reduction in mortality and the increasing proportion of patients who are discharged home.

Another observation based on our data was that 24-hr modified APACHE II scores (with the contribution of age and GCS removed) may have improved slightly over time, possibly implying that more recent patients receive more timely and complete resuscitation during the hours immediately following their neurological insult. Non-neurological organ dysfunction is an important predictor of outcomes in neurocritical care patients.66 Early physiological derangements are known to be associated with worse outcomes, and their timely correction could influence recovery.67-69 Consistent with this notion, one randomized controlled trial demonstrated that pre-hospital rapid sequence intubation (RSI) by paramedics increased the proportion of TBI patients with a favourable neurological recovery.14 Pre-hospital RSI was routine practice in Southern Alberta even before the time period assessed in this study. It is also possible that some of the improvements in outcomes that we observed in neurocritical care patients are attributable to advances in general critical care. However, another study assessing more than 25,000 consecutive patients admitted to ICUs in Calgary between 1999 and 2006 did not report any significant change in mortality.70

Our study is strengthened by the following factors: large cohort size (> 4,000 patients); long time period of interest (> 11 years); and prospective data collection by critical care physicians directly involved in patient care. Although not strictly a population-based study—we did not limit our cohort to Calgary residents—our study involved all consecutive patients with relevant diagnoses admitted to ICUs in Southern Alberta. Thus, our cohort is not subject to the referral biases of many single-center cohort studies and is a highly representative sample of neurocritical care patients within the population. The main limitation of our study is the lack of detailed long-term neurological outcome information. Discharge disposition data comprise an imperfect surrogate for neurological recovery. For example, patients who are transferred to another hospital may eventually be discharged home with a favourable recovery. In this study, such patients were characterized as having an unfavourable recovery. Conversely, some patients who are discharged home may remain severely disabled. The observation that an increasing number of neurocritical care patients in our region were eventually discharged home without the need for support services, however, suggests that the reduction in hospital mortality was not attributable simply to a greater proportion of survivors with a poor neurological recovery. Patients who survive in a profoundly disabled state are frequently unable to return home.

Conclusion

The proportion of critically ill neurological patients who survive and are discharged home has increased significantly in Southern Alberta over the past 11 years. These changes have been more pronounced in certain subgroups of patients, including those with TBI and SAH. Improved outcomes cannot necessarily be attributed to any one specific intervention. Our data suggest that contributing factors may include improvements in early resuscitation; availability of individuals with subspecialty expertise in neurocritical care; more attention to the prevention and expeditious treatment of factors that may contribute to secondary neurological injury; enhanced communication between critical care physicians and neuroscience specialists; increased use of standardized protocols; and changes in practices related to withdrawal of life-sustaining interventions. Implementation of some of these practices could potentially produce comparable improvements in outcomes in other regions. Further research is required to clarify which factors have the greatest impact on improving outcomes in neurocritical care patients and whether similar trends have occurred in other parts of Canada and the world.