Prevalence and Risk Factors for Intensive Care Unit Delirium after Traumatic Brain Injury: A Retrospective Cohort Study

Abstract

Background:

Delirium remains understudied after traumatic brain injury (TBI). We sought to identify independent predictors of delirium among Intensive Care Unit (ICU) patients with TBI.

Methods:

This single-center retrospective cohort study evaluated adult TBI patients requiring ICU admission. Outcomes included delirium days within the first 14 days, as assessed by the Confusion Assessment Method-ICU (CAM-ICU). Models adjusted for age, sex, insurance, Marshall head computed tomography (CT) classification, presence of subarachnoid hemorrhage (SAH), Injury Severity Score (ISS), need for cardiopulmonary resuscitation, maximum admission Glasgow Coma motor score, glucose, hemoglobin, and pupil reactivity.

Results:

Delirium prevalence was 60% with a median duration of 4 days (interquartile range [IQR]: 2-8) among ICU patients with TBI (n=2664). Older age, higher ISS, maximum motor score <6, Marshall Class II-IV, and SAH were associated with risk of increased delirium duration (all P<0.001).

Conclusions:

In this large cohort, ICU delirium after TBI affected 3 out of 5 patients for a median duration of 4 days. Age, general injury severity, motor score, and features of intracranial hemorrhage were predictive of more TBI-associated delirium days. Given the high prevalence and impact on hospitalization, further work is needed to understand the impact of delirium and TBI on outcomes, and if delirium risk can be minimized.

Introduction

Delirium is a temporary state of altered consciousness and cognition that is common among severely ill and hospitalized patients. The prevalence of delirium is estimated to be 35-67% in hospitalized trauma patients and as high as 71% in those admitted to intensive care units (ICU).(1–3) Delirium is associated with prolonged hospital and ICU length of stay, higher costs, increased morbidity, and a two-fold increased mortality.(4, 5) Additionally, in-hospital delirium is associated with worse cognitive function for up to a year after illness.(6–9) A recent systematic review found neurocritically ill patients are also subject to the deleterious effects of delirium, as its presence correlates with longer hospital stays, reduced functional independence, and poorer cognition.(10)

Given its fluctuating course and variable presentation, delirium may be overlooked when systematic screening methodologies are not adopted.(2) The difficulty in detecting delirium is greater still in the trauma population, where altered consciousness resulting from traumatic brain injury (TBI) increases the complexity of the diagnosis. The Confusion Assessment Method-ICU (CAM-ICU) is a commonly used, validated tool for diagnosing delirium in the critical care setting.(11) The reliability of the CAM-ICU for delirium assessment in patients with mild and moderate TBI is comparable to other current screening methodologies.(12, 13)

A number of modifiable and non-modifiable risk factors for delirium have been identified. Much of this work has focused on medical and surgical patients, and specifically excluded individuals with known neurocognitive pathologies. Delirium risk factors specific to the trauma population include advanced age, higher Acute Physiologic Assessment and Chronic Health Evaluation (APACHE) II score, higher Injury Severity Score (ISS), recent drug or alcohol use, cumulative doses of opiates and benzodiazepines, longer duration of mechanical ventilation, number of surgical procedures, and higher transfusion requirement.(3, 12, 14–16) While delirium is common in patients with TBI, risk factors for delirium development among severely injured trauma patients with TBI remain underexplored.(12, 16)

The purpose of this study was to identify independent predictors of delirium onset and duration in critically ill trauma patients with TBI utilizing delirium assessments validated in TBI. (12, 13) We hypothesized that admission characteristics, injury patterns, and physiologic variables would identify individuals at increased risk for the development of delirium during their index admission.

Methods

All methods were approved by the Institutional Review Board at Vanderbilt University Medical Center.

Study population

This single-center, retrospective cohort study included patients 16 and older admitted to Vanderbilt University Medical Center over a six-year period with acute TBI and an ICU stay greater than 24 hours, with robust linkage to Vanderbilt’s comprehensive medical record. TBI was classified based on International Classification of Diseases, Ninth Revision diagnostic codes, and included concussion without intracranial hemorrhage. Exclusion criteria included penetrating mechanism and missing admission patient data.

Data collection

Query of the medical record yielded admission demographics (e.g., age, sex, insurance status as classified as private, public, workers compensation, self-pay or none), injury-specific data (ISS), admission markers of TBI severity (Marshall Head CT classification, presence of subarachnoid hemorrhage (SAH), pupil reactivity, and maximum Glasgow Coma motor score),(17) and admission physiologic measurements (minimum glucose, minimum hemoglobin). These variables were a priori selected based on their demonstrated effect on TBI outcome in the prognostic model developed by the International Mission for Prognosis and Analysis of Clinical Trials in TBI (IMPACT).(17)

The primary outcome was delirium duration within the first 14 days of admission. The secondary outcome was delirium and coma-free days (DCFD) within the first 14 days of admission. Delirium measurement was based on daily assessments of the patient’s Richmond Agitation and Sedation Scale (RASS) and the CAM-ICU, both of which are performed at regular intervals by the bedside nurse as standard practice in our ICUs. RASS has high reliability and validity in critically ill adults.(18) Classification of delirium and coma was performed post-hoc as follows: (a) normal, RASS score of ≥ −3 and negative CAM-ICU; (b) delirious, RASS score of ≥ −3 and positive CAM-ICU; or (c) coma, RASS score of ≤ −4. Because the duration of delirium could be biased by the relationship between mortality and delirium, we evaluated this outcome only among those surviving their hospital stay (n=2135). The entire cohort (n=2664) was included in the analysis of DCFD.

Patients missing assessments tended to be older, less severely injured, have higher motor Glasgow coma scores, and received fewer opiates and benzodiazepines. Missing variables were assumed to be missing at random and handled with multiple imputation to reduce bias and account for uncertainty of predictions.

Statistical Models

Before modeling each outcome (i.e., primary outcome of delirium days and secondary outcome of delirium/coma free days), the variables were assessed for redundancy, which would indicate a high degree of correlation between the variables, of which none was detected. We used a multivariate negative binomial regression model to evaluate the relationship between delirium duration and admission risk factors. An incidence rate ratio (IRR) above 1.0 indicates increased rate of delirium days. The non-parametric distribution for DCFD led us to select a proportional odds logistic regression model. An odds ratio (OR) above 1.0 denotes greater odds of more delirium and coma-free days. For both analyses, the reference value for continuous variables (age, ISS, maximum motor score, glucose, hemoglobin) was set at the 25^th percentile, while the comparison value was the 75^th percentile for the respective cohort. This approach permits meaningful interpretation without the assumption of a linear relationship. Categorical variables were compared against a set reference value, which were as follows: pupils, both reactive; insurance type, private; and Marshall Head CT classification, grade I. The level of significance was set at P ≤ 0.05. Statistical analyses were performed using R version 3.3.1.

Results

Cohort description

Eligibility criteria yielded a final cohort of 2664 patients, with 2135 (80.1%) surviving until discharge (Figure 1 and Table 1). The cohort was 71% male (n=1892), with a median age of 42 years (interquartile range [IQR]: 25-59), and a median ISS of 27 (IQR: 20-36), indicating a severely injured polytrauma cohort. In 23% of patients (n=606), no intracranial pathology was noted on imaging (Marshall Class I), indicating a concussion and traumatic brain injury without hemorrhage. Half of the patients (n=1334) had evidence of SAH, while 7% (n=197) had a component of epidural bleed.

Figure 1.

Characteristics of analyzed cohorts. ICU, intensive care unit; TBI, traumatic brain injury; DCFD, delirium- and coma-free days

Table 1:

Patient demographics, admission injury characteristics, and hospital course for TBI admitted to the trauma ICU.

	All eligible patients (n=2664)
Admission data	median (IQR) or n (%)
Age at admission (years)	42 (25; 59)
Male gender	1892 (71)
Race
White	2297 (86)
Black	227 (9)
Hispanic	108 (4)
Other	32 (1)
Insurance type
Private	1127 (42)
Public	931 (35)
Self-Pay/None	516 (19)
Workers Compensation	90 (3)
Admission injury characteristics
Injury Severity Score (ISS)	27 (20; 36)
CPR Performed	89 (3)
Marshall Class
I	606 (23)
II	1481 (56)
III	256 (10)
IV	31 (1)
V/VI	290 (11)
Cerebral subarachnoid hemorrhage (SAH)	1334 (50)
Cerebral extra/epidural mass	197 (7)
Blunt injury	2664 (100)
In-hospital
Mental status assessed at least once	2479 (93)
Days with mental status info (per patient)	6.0 (3.0; 12.0)
Delirium
At least one day of delirium	1494 (60)
Days of delirium	4.0 (2.0; 8.0)
Coma	1511 (61)
At least one day of coma	1511 (61)
Days of coma	2.0 (1.0, 3.0)
Ventilator days	2.0 (1.0; 6.0)
Intensive care unit (ICU) stay (days)	4.0 (2.0; 8.0)
Hospital stay (days)	7 (3; 14)

This table includes describes the characteristics of a critically ill retrospective cohort with traumatic brain injury (N=2664).

TBI, traumatic brain injury; ICU, intenstive care unit; IQR, interquartile range; CPR, cardiopulmonary resuscitation.

Mental status assessments were documented at least once in 93% (n=2479) of the population. The median number of total days assessed per patient was 6 (IQR: 3-12 days). Delirium was identified in 60% (n=1494) of the full cohort and in 71% of those who survived to discharge. The median duration of delirium was 4 days (IQR: 2-8). At least one day of coma was present in 61% (n=1511) of this population, with a median duration of 2 days (IQR: 1-3). The median ICU length of stay was 4 days (IQR: 2-8), and the median number of days on a ventilator was 2 (IQR: 1-6). Of patients with CT-visible structural brain injury (Marshall Class II, III, or IV), 48% experienced delirium on at least one hospital day, whereas among patients without CT-visible TBI (Marshall Class I), only 30% experienced delirium on one hospital day. A diagram demonstrating possible transitions patients can undergo in this study is shown in Figure 2, with number of patients in each brain state on each hospital day shown in Figure 3.

Figure 2.

Possible transitions between brain states in the study population. ICU, intensive care unit

Figure 3.

Number of patients in transitional brain states (coma, delirium, normal) per day up to hospital day 30. Transitions often occur daily as individual patients move in and out of delirium, coma, and normal brain states before reaching permanent states of death or discharge.

Delirium days

All 2135 hospital survivors were included in the model for delirium duration, and 279 (13%) had missing data requiring imputation. Hospital length of stay less than 14 days accounted for the majority (10%) of missing data. The full results of the model are provided in Table 2. Older age (incidence rate ratio [IRR] 1.37; p-value <0.001), Marshall Class II, III, or IV (p-value <0.001), presence of SAH (IRR 1.42; p-value <0.001), and higher ISS (IRR 1.81; p-value <0.001) were all associated with greater risk for more days with delirium. Best Glasgow Coma motor score of 6 was associated with significantly fewer delirium days (IRR 0.92; p-value <0.001). Female gender may be protective against delirium (IRR 0.86; 95% CI: 0.74-0.98); however, this relationship was not statistically significant (p-value 0.067).

Table 2:

Multivariate negative binomial regression model of admission variables predicting delirium days for patients with TBI

	Reference	Comparison	IRR (95% CI)	P-value
Age (years)*	24.1	56.8	1.37 (1.22; 1.53)	<0.001
Gender	Male	Female	0.86 (0.74; 0.98)	0.067
Insurance type	Private	Public	1.06 (0.93; 1.21)	0.091
		Self-pay	0.84 (0.71; 1.00)
		Workers’ compensation	0.89 (0.65; 1.22)
Marshall head CT class	I	II	1.29 (1.09; 1.52)	<0.001
		III/IV	1.65 (1.28; 2.13)
		V/VI	0.86 (0.65; 1.13)
Subarachnoid hemorrhage (SAH)	No	Yes	1.42 (1.24; 1.62)	<0.001
Injury Severity Score (ISS)*	17	34	1.81 (1.58; 2.07)	<0.001
CPR performed	No	Yes	1.70 (1.01; 2.86)	0.060
Max motor score*^	5 or lower	6	0.92 (0.88; 0.95)	<0.001
Pupil reactivity^	Both reactive	One reactive	1.25 (0.99; 1.59)	0.076
		Both fixed	0.85 (0.66; 1.09)
Min glucose (mmol/L)*^	97	136	0.93 (0.86; 1.01)	0.185
Min hemoglobin (g/dL)*^	10	13.3	0.88 (0.80; 0.96)	1.000

This table includes results from hospital survivors (n=2135) in a retrospective cohort of patients 16 years and older admitted to the trauma intensive care unit with traumatic brain injury.

^*Reference group is 25^th percentile, comparison group is 75^th percentile.

^{^}Day of admission.

TBI, traumatic brain injury; IRR, incidence rate ratio; CT, computed tomography; CPR, cardiopulmonary resuscitation.

Delirium/coma free days

All patients in the final cohort (n=2664) were included in the model for DCFD. Missing data was present in 16% (n=435), the majority due to a hospital stay shorter than 14 days. Older age (OR 0.75; p-value <0.001), higher ISS (OR 0.36; p-value <0.001), Marshall Class II, III, or IV (p-value <0.001), and presence of SAH (OR 0.64; p-value <0.001) were associated with decreased odds of DCFD (Table 3). Higher minimum glucose (OR 1.14; p-value 0.005), higher minimum hemoglobin (OR 1.34; p-value <0.001), motor score of 6 (OR 1.15; p-value <0.001), and female gender (OR 1.25; p-value 0.007) correlated with increased days without delirium or coma. One reactive pupil on the day of admission was associated with fewer DCFDs, while both fixed pupils predicted more DCFDs (“normal” days). Relative to individuals with private insurance, patients with public insurance had fewer DCFD (OR 0.84; 95% CI: 0.71-0.99) and those with no coverage (self-pay) had more DCFD (OR 1.46; 95% CI: 1.19-1.78).

Table 3:

Proportional odds multivariable logistic regression of admission variables predicting delirium- and coma-free days

	Reference	Comparison	Odds ratio (95% CI)	P-value
Age (years)*	24.1	56.8	0.75 (0.65; 0.88)	<0.001
Gender	Male	Female	1.25 (1.06; 1.46)	0.007
Insurance type	Private	Public	0.84 (0.71; 0.99)	<0.001
		Self-pay	1.46 (1.19; 1.78)
		Workers’ compensation	1.13 (0.76; 1.68)
Marshall head CT class	I	II	0.68 (0.55; 0.84)	<0.001
		III/IV	0.59 (0.44; 0.79)
		V/VI	1.04 (0.77; 1.39)
Subarachnoid hemorrhage (SAH)	No	Yes	0.64 (0.54; 0.75)	<0.001
Injury Severity Score (ISS)*	17	34	0.36 (0.31; 0.42)	<0.001
CPR performed	No	Yes	1.10 (0.75; 1.61)	0.641
Max motor score*^	5 or lower	6	1.15 (1.10; 1.21)	<0.001
Pupil reactivity^	Both reactive	One reactive	0.67 (0.49; 0.91)	<0.001
		Both fixed	3.00 (2.32; 3.88)
Min glucose (mmoL/L)*^	97	136	1.14 (1.03; 1.25)	0.005
Min hemoglobin (g/dL)*^	10	13.3	1.34 (1.19; 1.52)	<0.001

This table includes results from a retrospective cohort (N=2664) of patients 16 years and older admitted to the trauma intensive care unit with traumatic brain injury.

^*Reference group is 25^th percentile, comparison group is 75^th percentile.

^{^}Day of admission.

CT, computed tomography; CPR, cardiopulmonary resuscitation.

Discussion

This study evaluates a large retrospective cohort of critically ill trauma patients with TBI for independent predictors of delirium onset and duration within the first 14 days of admission. Delirium developed in 60% of this population, a finding consistent with previously published studies.(1–3) In our cohort, increased age, severe injury, presence of intracranial findings on imaging, and reduced Glasgow Coma motor score were significantly associated with both increased delirium duration and fewer days with normal cognition (i.e., DCFD). Pathophysiologically, these findings demonstrate age and greater TBI severity are associated with delirium. Higher minimum glucose and hemoglobin levels on hospital day one, as well as female gender, appear to be protective against delirium in this population. Pathophysiologically, hypoglycemia and hypotension (a major cause in trauma patients being hemorrhage) can cause secondary brain injury,(19) which may be related to delirium in TBI patients. We see an effect of insurance type in one model and not the other, and thus we are cautious in interpreting this. However, it is likely that insurance status may be a proxy for socioeconomic status, which is likely a predictor of delirium in TBI as well as overall TBI outcome. The relationship of admission pupil reactivity to delirium is contradictory and present in only one statistical model. We believe this likely reflects issues with the reliability of admission pupil reactivity (i.e. an assessment occurring during a busy trauma resuscitation). This study is the largest yet undertaken to evaluate the prevalence and risk factors of delirium in critically-ill trauma patients with TBI, offering new insight into delirium risk in this unique population and serving as a first step towards reducing the incidence of this common condition.

Our understanding of delirium as a distinct neuropsychological phenomenon in the hospitalized patient has evolved rapidly in recent years. Improved diagnostic algorithms have led to the identification of delirium subtypes with significant disparities in short- and long-term cognitive outcomes. Recent work by Girard and colleagues identified four clinical phenotypes of delirium in the critically-ill: hypoxic, septic, sedative-associated, and metabolic.(3) Delirium attributable to metabolic derangements was associated with the best cognitive function at 3- and 12-month follow-up; however; no difference in mortality was detected between delirium subtypes. It is likely that the delirium observed in our study may be a fifth clinical phenotype that is distinct from those common in medical and surgical ICUs given its association with structural brain injury. This phenotype could be termed structural delirium, but further study would be required in non-traumatically brain injured patients (e.g., ischemic stroke, hemorrhagic stroke, etc) to clarify if this phenotype is related to structural brain injury or specific to TBI. Future research will need to consider the modifying effect of hypoxia, sepsis, sedatives, and/or metabolic derangement on structural delirium. We also recognize that the diagnosis of delirium in structurally brain injured patients remains controversial, as it is difficult to determine if delirium is the result of structural brain injury or systemic effects (which has been better characterized). However, given that delirium is a distinct phenotype with a definition in the Diagnostic and Statistical Manual of Mental Disorders, 5th Edition,(20) if delirium is measured in TBI or structurally brain injured patients, we argue that nuanced efforts should be made to better describe this phenomenon.

Early identification of patients at increased risk for delirium is essential. Prior studies have found the average time to delirium onset to be 3-5 days following hospital admission.(4, 16) No therapeutic intervention has yet shown to reliably reverse delirium once present, but judicious medication selection, environment optimization, and early mobilization strategies reduce the likelihood of its onset.(16, 21–26) Though such interventions remain understudied in the critically ill TBI population, a recent randomized trial demonstrated that selecting dexmedotomidine over haloperidol reduced delirium in this high-risk population.(23, 27) By identifying those at increased risk for delirium upon admission to the trauma ICU, the intensivist can initiate such interventions in a targeted fashion. Our findings help alert providers to vulnerable patients requiring extra consideration for delirium prevention and interventions.

Advances in care over the past decade have improved TBI survivorship, resulting in an estimated 5.3 million Americans living today with permanent TBI-related disability.(28) The core principles of post-TBI care are maintenance of adequate cerebral perfusion and avoidance of hypoxic injury to prevent secondary injury to threatened brain tissue.(19) Despite known associations between delirium and poor cognitive outcomes, post-injury delirium is not considered at present a preventable “second hit” by the Brain Trauma Foundation.(6–9, 19) Continued work in this field will improve our understanding of the long-term detrimental effects of delirium in the TBI patient and will help better identify viable preventative and treatment strategies. The anticipated effects of such efforts are fewer hospital and ICU days, reduced healthcare-associated costs, and improved long-term cognitive function of individuals in this sizable, vulnerable population.

Approximately 23% of patients had negative CT scans (Marshall Class I) but carried a diagnosis of TBI. This reflects an important category of patients, traditionally diagnosed with a concussion, without lesions on CT scan but in whom magnetic resonance imaging (MRI) or emerging biomarkers, such as glial fibrillary acidic protein (GFAP) or ubiquitin C-terminal hydrolase-L1 (UCH-L1), likely reveal underlying TBI pathology predictive of long-term neurologic outcome.(29–32) This is an important population to include in an analysis of the risk factors for delirium in TBI.

Limitations of our study are largely attributable to its retrospective and single-center nature. Missing data was addressed with multiple imputation; we feel this approach strengthened our modeling as compared to a complete case analysis. Daily neurologic assessments were performed by bedside nurses with standard critical care training, which increases the generalizability of these findings but may reduce accuracy and/or reliability. This reflects the health services-oriented approach of this study. We did not take into consideration any interventions provided prior to patient arrival or during the 14-day assessment period. While it is understood that certain interventions and/or complications are associated with higher rates of delirium, our primary goal was to identify admission patient characteristics associated with increased risk of delirium onset. Our group has an ongoing cohort (NIH R01 GM120484; NCT03098459)(33) that will not only be able to examine the impact of in-hospital interventions, such as medications, on delirium, but also track long-term neurocognitive outcomes for critically ill trauma patients with and without TBI. This cohort will also better illuminate the impact of medical co-morbidities on these relationships. Finally, there is ongoing uncertainty about the validity of CAM-ICU assessments in the severe TBI population, with a previous study demonstrating reliability of the CAM-ICU in a general critically ill trauma population, which included patients with severe TBI.(34)

Conclusions

To our knowledge, this is the first study aimed to identify risk factors for delirium in a critically ill cohort with TBI. Important delirium risk factors found were increased age, severity of injury, presence of intracranial findings on imaging, reduced motor score, lower glucose and hemoglobin levels, and male gender. The known detrimental effects of delirium (4–10) combined with its high prevalence in TBI warrant further investigation, for which we have ongoing projects.(33) Future work should strive to better characterize structural delirium and long-term impacts on the injured brain. Researchers should also seek to identify early interventions effective at preventing delirium onset and reducing its duration in this unique population.