Corticosteroids and transition to delirium in patients with acute lung injury

Abstract

Objective

Delirium is common in mechanically ventilated patients in the intensive care unit (ICU) and associated with short- and long-term morbidity and mortality. The use of systemic corticosteroids is also common in the ICU. Outside of the ICU setting, corticosteroids are a recognized risk factor for delirium, but their relationship with delirium in critically ill patients has not been fully evaluated. We hypothesized that systemic corticosteroid administration would be associated with a transition to delirium in mechanically ventilated patients with acute lung injury (ALI).

Design

Prospective cohort study

Setting

Thirteen ICUs in 4 hospitals in Baltimore, MD

Patients

520 mechanically ventilated adult patients with ALI

Measurements and Main Results

Delirium evaluation was performed by trained research staff using the validated CAM-ICU screening tool. A total of 330 (64%) of the 520 patients had at least two consecutive ICU days of observation in which delirium was assessable (e.g., patient was non-comatose), with a total of 2286 days of observation and a median (inter-quartile range [IQR]) of 15 (9, 28) observation days per patient. These 330 patients had 99 transitions into delirium from a prior non-delirious, non-comatose state. The probability of transitioning into delirium on any given day was 14%. Using multivariable Markov models with robust variance estimates, the following factors (adjusted odds ratio, 95% confidence interval) were independently associated with transition to delirium: older age (compared to <40 years old, 40-60 years (1.81, 1.26 to 2.62) and >=60 years (2.52, 1.65 to 3.87)) and administration of any systemic corticosteroid in the prior 24 hours (1.52, 1.05 to 2.21).

Conclusions

After adjusting for other risk factors, systemic corticosteroid administration is significantly associated with transitioning to delirium from a non-delirious state. The risk of delirium should be considered when deciding about the use of systemic corticosteroids in critically ill patients with ALI.

Delirium is common in mechanically ventilated patients in the intensive care unit (ICU) and associated with short- and long-term morbidity and mortality.[1-6] Delirium can be ameliorated through addressing modifiable risk factors, such as immobilization, the administration of benzodiazepines and opioids, and sleep disruption.[7-14] While delirium may involve neuroinflammation, systemic corticosteroids, despite their anti-inflammatory properties, are recognized as a risk factor for delirium and other psychiatric phenomena in hospitalized patients.[15-18] However, there has been little evaluation of the association between systemic corticosteroids and delirium in the ICU setting. Such investigation is important because corticosteroids are frequently used with evolving data regarding their potential benefits within the ICU setting.[19-24] Hence, further evaluation of the potential risks of corticosteroids[25] is warranted to assist with decision-making regarding the benefits and risks of their use.

Given their high severity of illness and frequent requirement for mechanical ventilation and sedation, patients with acute lung injury (ALI) are at especially high risk for delirium and also frequently receive corticosteroids.[14, 20, 21, 26-29] Hence, using a cohort of ALI patients, our objective was to evaluate if systemic corticosteroid use was an independent risk factor for transitioning to delirium from a noncomatose, non-delirious state.

METHODS

Patients

This project was conducted as a secondary analysis of a prospective cohort study (trial registration NCT00300248) of 520 consecutive, mechanically ventilated patients with ALI [30] recruited from 13 ICUs at four teaching hospitals in Baltimore, MD between October 2004 and October 2007. All study ICUs utilized goal directed sedation protocols and the two largest study sites utilized daily interruption of sedation infusions. Patients were excluded from the prospective cohort study if they met any of the following criteria: (1) preexisting comorbid illness with a life expectancy of six or less months (for example, metastatic cancer), (2) preexisting cognitive impairment or communication/language barriers, (3) no fixed address, (4) transferred from another hospital and had pre-existing acute lung injury for >24 hours, (5) mechanically ventilated for >5 days before onset of acute lung injury, (6) previous lung resection, and (7) a physician order for no escalation of care in the ICU (for example, no vasopressors) at ALI onset.[31]

Primary outcome: transition to delirium

The primary outcome was patients’ transition from a “normal” state (i.e., non-delirious and noncomatose) to a delirious state on the next day, as described herein. Patients’ sedation and delirium status was evaluated once-daily by rigorously trained research staff using the validated Richmond Agitation Sedation Scale (RASS) and Confusion Assessment Method for the ICU (CAM-ICU), respectively.[32, 33] On the basis of these assessments, on a daily basis, patients are classified into 1 of 3 mutually exclusive categories: (1) comatose (RASS of −4 or −5), (2) delirious (RASS >-3 with a positive CAM-ICU), or (3) normal (RASS >-3 with a negative CAM-ICU); this information was used to determine the primary outcome

Primary exposure: systemic corticosteroids

The primary exposure was receipt of systemic corticosteroids in the ICU, as obtained from medication administration records and modeled as daily administration status (yes/no) and total daily dose (in milligrams of prednisone-equivalents).[34]

Covariates: potential confounders

Based on prior literature, [1, 10, 11, 35-40] a number of potential confounding variables were considered in this analysis. In addition to patient demographics (age, sex, and race), the following baseline variables were evaluated using data collected from review of medical records: Charlson comorbidity index[35], prior cerebral vascular accident, current or prior excess alcohol use, current or prior illicit drug use, and prior home use (either as-needed or routine scheduled dosing) of any medication within each of the following drug classes: systemic corticosteroids, opioids, or benzodiazepines. The following ICU covariates also were considered: type of ICU (medical vs. surgical), Acute Physiology and Chronic Health Evaluation II (APACHE II)[41] severity of illness score at ICU admission, daily Sequential Organ Failure Assessment (SOFA) score[42] (excluding the neurologic component to prevent adjusting for a component of the primary outcome within the regression model), any use of continuous dialysis, daily presence of sepsis (evaluated based on pre-existing diagnostic criteria[43] and microbiologically documented infection), and daily administration status (yes/no) and total daily dose of opioids or benzodiazepines (presented as morphine- and midazolam-equivalent doses, respectively).[44-46] Propofol and dexmedetomidine were rarely used in the study-site hospitals and therefore not relevant covariates for this analysis.

Statistical Analysis

Descriptive statistics were calculated for continuous and categorical variables. To appropriately model continuous exposure variables (e.g. age), their individual associations with the probability of transitioning to delirium was evaluated through inspection of locally weighted least squares (LOWESS) plots.[47]

A first-order Markov model[48] was used to estimate the relative odds of transitioning from a normal state to a delirious state on the next day. This model uses logistic regression analysis to evaluate the primary exposure variable (systemic corticosteroids) against delirium (vs. normal state) on a specific day, adjusted for: (1) the presence of delirium on the immediately prior day, and (2) the previously described covariates. Robust variance estimates were used to account for correlation in the repeated daily measures of delirium status within patients beyond the first-order Markov assumption.[49] As done in prior studies,[36, 50] any days in which delirium status was not assessable (e.g., due to coma, subject refusal, or a missed assessment) or in which delirium status was not available on the immediately prior day (e.g., day of admission, coma, subject refusal, or missed assessment) were not included in the analysis in order to restrict the analysis to consecutive days of completed delirium assessments.

Among all of the covariates considered for this analysis (as described previously), individual covariates were considered for inclusion in the multivariable regression model if they had an association, at p ≤0.2, with the primary outcome variable based on bivariable logistic regression analyses. In the final multivariable regression model, the absence of multicolinearity was confirmed by evaluating variance inflation factors[49] and goodness-of-fit testing was confirmed with the Hosmer-Lemeshow statistic.[49] All statistical analyses were performed using Stata version 12 (Stata Corp., College Station, TX, USA). The study was approved by institutional review boards at all participating sites.

Results

A total of 330 (64%) of the 520 consecutive, mechanically ventilated ALI patients had at least 2 consecutive days that were assessable for delirium, representing 2286 days of observation, with a median (inter-quartile range [IQR]) of 15 (1, 28) days of observation per patient. The 330 patients had a median (IQR) age of 51 (41, 63) years, with 55% male and 60% Caucasian (Table 1). Patients’ median (IQR) APACHE II score was 24 (19, 30), with 70% of patients admitted to a medical ICU (MICU) (Table 1) and 37% of patients receiving systemic corticosteroids.

Table 1.

Patient Characteristics (n=330)

Characteristic	Result
Demographics
Age, median (IQR) years	51 (41, 63)
Male, No. (%)	183 (55)
Caucasian race, No. (%)	199 (60)
Charlson comorbidity index score, median (IQR)	2 (1, 4)
Comorbidities, No. (%)
Current or prior illicit drug use	87 (26)
Current or prior excess alcohol use	79 (24)
Prior cerebral vascular accident	29 (9)
Home Medication Use, No. (%)
Opioids	70 (21)
Systemic corticosteroids	45 (14)
Benzodiazepines	29 (9)
Critical Illness
Medical ICU admission, No. (%)	231 (70)
APACHE II score, median (IQR)	24 (19, 30)
Ever use of continuous hemodialysis, No. (%)	29 (9)

Abbreviations: IQR – Inter-quartile Range, ICU – Intensive Care Unit, APACHE II – Acute Physiology and Chronic Health Evaluation II[41] score

Delirium was documented on one or more days in 83% of patients with a median (IQR) duration of delirium of 7 (3, 12) days. Among 733 previous days with an observed normal state, there were 99 transitions to delirium (14% daily probability of transition). Corticosteroids were administered on the prior day in 41% and 34% of days with and without a transition to delirium, respectively (Table 2). Additionally, a transition to delirium was present in 16% and 12% of days with and without administration of corticosteroids respectively. Among days in which corticosteroids were received, the median (IQR) dose in those with and without a transition to delirium was 44 (13, 75) and 23 (13, 50) milligrams of prednisone-equivalent respectively.

Table 2.

Summary of Daily ICU Variables, by Delirium and Corticosteroid Status^a

Variable	“Normal” (Delirium/ Coma-Free), n = 634	Delirium, n = 99	No Corticosteroid Administered, n =478	Corticosteroid Administered , n = 255
Daily SOFA scoreb, median (IQR)	5 (3, 8)	6 (4, 9)	5 (2, 7)	7 (4, 12)
Daily Sepsis status, No. (%)	412 (65)	62 (63)	283 (59)	191 (75)
Use of corticosteroids in prior 24 hours, No. (%)	214 (34)	41 (41)	-	-
Prednisone-equivalent dose (if any use), median (IQR) mg	23 (13, 50)	44 (13, 75)	-	-
Use of benzodiazepine in prior 24 hours, No. (%)	183 (29)	33 (33)	164 (34)	52 (20)
Midazolam-equivalent dose (if any use), median (IQR) mg	8 (3, 23)	10 (5, 28)	7 (3, 17)	14 (5, 39)
Use of narcotic in prior 24 hours, No. (%)	372 (57)	60 (61)	274 (57)	158 (62)
Morphine-equivalent dose (if any use), median (IQR) mg	22 (7, 65)	47 (9, 160)	27 (8, 84)	20 (5, 67)
Presence of delirium, No. (%)	-	-	58 (12)	41 (16)

Abbreviations: SOFA – Sequential Organ Failure Assessment Score[42] IQR – Inter-quartile Range,

^aData are presented for 330 patients with 733 patient days that were included in the analysis of transition to delirium (i.e. “normal” status on previous day). The daily probability of transition to delirium was 14% (99/733). Data for administration of corticosteroids are presented to reflect administration on the 733 days included in this assessment.

^bSOFA score[42] is modified to exclude the neurologic component to prevent adjusting for a component of the primary outcome within the regression model

Results of the bivariable and multivariable logistic regression models are presented in Table 3. In the multivariable model, the administration of systemic corticosteroids was significantly associated with transition to delirium (adjusted Odds Ratio [OR] = 1.52, 95% Confidence Interval [CI] = 1.05 - 2.21); however, in those receiving corticosteroids, there was no significant association between dose and delirium transition. Additionally, older age (compared to <40 years old, 40-60 years (adjusted OR: 1.81, 95% CI: 1.26 to 2.62) and >=60 years (2.52, 1.65 to 3.87)) was independently associated with delirium transition.

Table 3.

Patient and ICU variables associated with transition to delirium

	Odds Ratio (95% CI) of Transitioning to Deliriuma
Variable	Bivariable Analysis	Pb	Multivariable Analysis	Pb
Patient characteristics
Age <40 years old	Ref		Ref
Age 40- 60 years old	1.65 (1.17, 2.32)	< 0.01	1.81 (1.26, 2.62)	0.002
Age >60 years old	2.18 (1.37, 3.47)	< 0.001	2.52 (1.65, 3.87)	< 0.001
Male	1.31 (0.94, 1.82)	0.11	1.34 (0.96, 1.86)	0.08
Caucasian race	0.79 (0.57, 1.10)	0.17	0.74 (0.53, 1.02)	0.06
Charlson comorbidity index score	1.04 (0.97, 1.11)	0.25
Current or prior illicit drug abuse	1.16 (0.64, 2.10)	0.62
Current or prior excess alcohol use	1.16 (0.81, 1.67)	0.42
Prior cerebral vascular accident	1.29 (0.71, 2.37)	0.41
Home use of Opioids	1.15 (1.01, 1.31)	0.04	1.11 (0.97, 1.27)	0.12
Home use of systemic corticosteroids	0.89 (0.72, 1.12)	0.32
Home use of benzodiazepines	1.04 (0.84, 1.28)	0.74
ICU-related characteristics
Medical ICU admission	0.87 (0.57, 1.31)	0.49
APACHE II score at ICU admission	1.01 (1.00, 1.03)	0.10	1.01 (1.00, 1.03)	0.31
Use of continuous hemodialysis, ever	1.08 (0.69, 1.71)	0.73
Daily SOFA scorec	1.05 (1.01, 1.09)	0.02	1.03 (0.99, 1.07)	0.11
Daily sepsis status	1.25 (0.93, 1.67)	0.14	1.06 (0.79, 1.41)	0.71
Corticosteroid Administration	1.48 (1.05, 2.08)	0.03	1.52 (1.05, 2.21)	0.03
Higher Corticosteroid dose (per 40mg prednisone equivalent in 24h)	1.03 (0.94, 1.13)	0.56	0.97 (0.89, 1.07)	0.57
Benzodiazepine Administration	1.33 (1.01, 1.75)	0.05	1.32 (0.93, 1.89)	0.12
Higher Benzodiazepine dose (per 5mg midazolam equivalent in 24h)	1.02 (0.99, 1.04)	0.13	1.02 (0.99, 1.04)	0.18
Opioid Administration	1.16 (0.88, 1.52)	0.29

Abbreviations: CI, confidence interval, ICU – Intensive Care Unit, APACHE II – Acute Physiology and Chronic Health Evaluation II[41] score, SOFA, Sequential Organ Failure Assessment score[42]

^aOdds Ratios are interpreted as the odds of a transition to a delirious state from a normal state given the characteristic (for binary exposures) or given a one unit increase (for continuous exposures)

^bp values calculated using multivariate logistic regression adjusted for delirium status on the previous day, with robust variance estimates for repeated measures on the same subject.

^cSOFA score[42] is modified to exclude the neurologic component to prevent adjusting for a component of the primary outcome within the regression model

Because of a potential benefit from the anti-inflammatory effect of systemic corticosteroids on neuroinflammation associated with delirium, post-hoc sensitivity analyses were conducted to evaluate if early use of corticosteroids had a protective effect against transition to delirium. The following multivariable analyses demonstrated no association between early use of corticosteroids and transition to delirium: (1) evaluation of corticosteroids administered on ICU day 1 (p=0.63), (2) evaluation of corticosteroids administered on ICU day 1 or as an outpatient prior to admission (p=0.91), and (3) evaluation of corticosteroids administered on ICU day 1 only among those receiving steroids prior to admission (p=0.2). Furthermore, assessment of corticosteroid administration in terms of glucocorticoid-equivalents demonstrated no material difference from our primary multivariable results.

Discussion

In this prospective, multi-site cohort study evaluating 330 ALI patients with 2286 ICU days of observation, there is a significantly independent association between systemic corticosteroid use and transition to delirium the following day. Older age was also independently associated with transition to delirium.

Systemic corticosteroids are frequently used in the ICU [19-24] with 37% of ALI patients in this cohort having received them. However, use of systemic corticosteroids in the ICU is variable with evolving evidence regarding their use in critically ill patients; [19-24] thus, motivating the need to better understand their potential risks and benefits in this setting.

Our findings are consistent with prior studies of corticosteroids and delirium conducted outside of the ICU setting. For example, Kenna et.al.[16] reported on a case series of 55 patients and describe a range of neuropsychiatric outcomes, including delirium, occurring after steroid administration. Furthermore, in a multicenter cohort study of patients receiving >20 mg/day of prednisone-equivalents, Fardet et.al.[51, 52] observed that 52.5% of patients developed neuropsychiatric complications including delirium. A retrospective analysis of primary care patients also reported an independent association between corticosteroid administration and delirium, confusion or disorientation (hazard ratio 5.14, 95% Confidence Interval 4.54 - 5.82). Additionally, in a trial specifically evaluating delirium in a mixed-setting postoperative population adjusted for use of ICU care, an association with systemic corticosteroids has been described.[53]

While our study reveals an important relationship between corticosteroid administration (as a binary variable) and transition to delirium, we could not detect a significant dose-relationship. Our study identified a wide range of corticosteroid doses. However, of the 898 days on which a corticosteroid was administered, 72% of these days were at or above the dosing threshold previously associated with neuropsychiatric complications [51, 52] and this may have precluded our identifying a dose-response relationship.

Compared to this prior research, our study has the advantage of prospectively evaluating, on a daily basis, transition to delirium, as evaluated by trained research staff using a validated screening instrument (CAM-ICU), and adjusting for potential confounding risk factors. Our study builds on prior assessments of delirium in mechanically ventilated patients in the ICU.[36, 54-57] Unlike prior ICU studies assessing systemic corticosteroids and delirium,[54, 58, 59] our analysis demonstrated a statistically significant independent relationship, likely attributable to greater sample size, more frequent corticosteroid administration, and patient population that was at high risk for delirium.

Several studies [36, 56, 60, 61] have demonstrated a significant association of benzodiazepines with transitioning to delirium. Our study also identified a positive relationship between benzodiazepine administration and transition to delirium, but this relationship was not statistically significant in our multivariable model, perhaps because we adjusted for covariates, such as daily SOFA score and corticosteroids, which were not always considered in prior studies.[36, 56, 60, 61] Others have also identified a positive, but not statistically significant, relationship of benzodiazepine and delirium, but did identify a significant association with coma.[62]

Our study also confirms age as an important predictor of transitioning to delirium.[36, 58, 59] Particularly in older patients, attention should be directed towards addressing modifiable risk factors for delirium (e.g., use of corticosteroids and immobilization [7-9]).

Our study has potential limitations. First, as an observational study, it cannot prove a cause-effect relationship between systemic corticosteroids and delirium, and residual confounding may contribute to our findings. However, a randomized trial designed to specifically address this relationship between corticosteroids and delirium is not ethically possible. Hence, we encourage investigators in all future randomized, placebo-controlled studies of corticosteroids to include a safety outcome related to daily delirium status as well as secondary outcomes related to long-term cognitive and psychiatric outcomes[5, 63] to more fully evaluate the potential neuropsychological risks and benefits of corticosteroids in the critically ill. Second, we were unable to evaluate all known risk factors for delirium. In particular, dementia could not be evaluated since it was an exclusion criterion for our study and data regarding restraint use was unavailable. However, the exclusion of dementia helped reduce potential measurement error in our primary outcome since delirium assessment in critically ill demented patients is challenging. Additionally, we were unable to evaluate the effects of propofol[64] and dexmedetomidine[60, 65] since they were rarely used in the study hospitals. Our study did not have the ability to adjust for the indication for usage of systemic corticosteroids, and as previously noted the percentage of patients (72%) receiving clinically important doses of corticosteroids may have precluded our ability to identify a dose-response. Furthermore, we were unable to assess the association between delirium and mortality as there were only 15 deaths in the entire study population in which delirium status was available on the previous day with none of these occurring on a day following a non-comatose, non-delirious state (the structure of our Markov model). Finally, there may be limitations regarding the generalizability of our results because only ALI patients at teaching hospitals were studied with once daily assessments of delirium. However, because our patient population included a diverse range of patient demographics and comorbidities, both medical and surgical ICUs, and multiple hospitals, we believe our study is a valuable contribution to prior single site studies.

In conclusion, in this multi-site study of 330 mechanically ventilated patients with ALI, there was a significant association between the administration of systemic corticosteroids and transition to delirium, even after adjustment for use of sedative medications, relevant patient-specific risk factors, and severity of illness measures. Given the known short- and long-term morbidity and mortality associated with delirium in the ICU, these findings should be considered when weighing potential risk and benefits of using systemic corticosteroids in patients with ALI.