Pharmacologic Management of Intensive Care Unit Delirium: Clinical Prescribing Practices and Outcomes in More Than 8500 Patient Encounters : Anesthesia & Analgesia

Secondary Logo

Journal Logo

Original Research Articles: Original Clinical Research Report

Pharmacologic Management of Intensive Care Unit Delirium: Clinical Prescribing Practices and Outcomes in More Than 8500 Patient Encounters

Boncyk, Christina S. MD*,†; Farrin, Emily MD*; Stollings, Joanna L. PharmD†,‡; Rumbaugh, Kelli PharmD; Wilson, Jo Ellen MD, MPH†,§; Marshall, Matt PharmD; Feng, Xiaoke MS; Shotwell, Matthew S. PhD; Pandharipande, Pratik P. MD, MSCI*,†; Hughes, Christopher G. MD, MS*,†

Author Information
doi: 10.1213/ANE.0000000000005365

Abstract

KEY POINTS

  • Question: What are the prescribing practices for treatment of intensive care unit (ICU) delirium, and do medication choices impact in-hospital patient outcomes?
  • Findings: Antipsychotic medications are frequently utilized for treatment of ICU delirium, initiated early in the course, continued in more than 1 in 5 patients up to hospital discharge, and associated with key in-hospital outcomes.
  • Meaning: Antipsychotic medications may not portend benefit in treatment of ICU delirium, may introduce additional harm, and use should be evaluated at transitions of care to prevent inadvertent continuation.

Delirium is a disturbance of consciousness with fluctuation in mental status and changes in cognition that develop during a short period of time due to a medical condition, medication, intoxicating substance, or combination of causes.1 The physical and pharmacologic derangements common in critically ill patients lead to a high prevalence of delirium in intensive care units (ICUs).2 Decreasing the incidence and duration of delirium is paramount as delirium is independently associated with worse outcomes, including increased mortality and cognitive impairment.3,4

Nonpharmacologic and pharmacologic strategies have been shown to decrease the incidence of delirium.5,6 Once delirium develops, however, treatment is generally supportive with limited evidence-based pharmacologic treatment options. Antipsychotic medications, oral alpha-2 agonists (eg, guanfacine), and anticonvulsant medications (eg, valproic acid) have emerged as a potential pharmacologic option after results from initial studies suggesting their use could decrease delirium duration and/or severity.7–10 Recent larger randomized studies and systematic reviews, though, have failed to show a benefit of antipsychotic medications or oral alpha-2 agonists in decreasing the incidence or duration of delirium,11–13 and their routine use for treatment of delirium is not recommended under current practice guidelines.14 These medications are frequently prescribed in clinical practice for the management of ICU delirium, particularly hyperactive symptoms, to control patient agitation or assist in patient care,15 yet the prescribing patterns, efficacy, and the associated outcomes with prescribing these medications are not known.

We aimed to identify prescribing practices for treatment of delirium across multiple ICUs in a large tertiary academic center and to investigate the independent associations of antipsychotic medications, guanfacine, and valproic acid administration on delirium resolution, in-hospital mortality, days alive and free of the ICU (ICU-free days), and days alive and free of the hospital (hospital-free days) in patients with ICU delirium.

METHODS

Study Design and Participants

Our study was approved by local institutional review board with requirement for written informed consent waived by virtue of the retrospective audit using deidentified data sets (no. 171671) and this article adheres to applicable Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines. We performed a retrospective review of the electronic health record (EHR) for adult patients (≥18 years) admitted to Vanderbilt University Medical Center medical, surgical, trauma, or cardiovascular ICUs from January 1, 2014, to December 31, 2016. We included all patient encounters with delirium in the ICU as defined by at least 1 positive assessment on the Confusion Assessment Method for the ICU (CAM-ICU)16 while in the ICU. Data collection for participants was initiated on ICU admission to capture those patients who developed delirium within the ICU as well as investigate those patients who had delirium on ICU admission. We excluded patients with home antipsychotic prescriptions. Standard of care practices at our institution during this timeframe required sedation to be assessed by the Richmond Agitation-Sedation Scale (RASS) every 4 hours and delirium monitored with the CAM-ICU at least every 12 hours, though in practice our bedside nurses reliably assess both on average every 4 hours.17 Delirium management strategies in our ICU are centered on the “ABCDEF” bundle5,18 and using medications only for symptom control of hyperactive (agitated) delirium. The ABCDEF bundle incorporates evidence-based strategies shown to improve patient outcomes and recovery using the following principles: Assessing and treating pain, Both awakening and breathing trials, Choice of appropriate sedation, Delirium monitoring and management, Early mobility and exercise, and Family engagement.18 Examples of nonpharmacologic measures at our center include reorientation, provision of glasses and hearing aids, sleep hygiene, removal of catheters and restraints, and mobilization.

Procedures

We collected baseline patient demographic information, admission characteristics, including patient severity of illness, and daily hospital course data. Baseline characteristics collected on ICU admission included age, sex, body mass index (BMI), and home medications (antipsychotics, opioids, antidepressants, and benzodiazepines). Severity of illness was assessed using the University Health System Consortium (UHC)–expected mortality index, a validated model constructed using administrative data including diagnoses, comorbidities, and procedural information to predict mortality among hospitalized patients19–21 and shock diagnosis (defined as lactate >2 mmol/L with norepinephrine administration). Hospital course data included CAM-ICU assessments, RASS scores, medication administration, date of ICU admission, initial ICU location (trauma, surgical, medical, or cardiovascular ICU), days of mechanical ventilation, date of ICU discharge, date of hospital discharge, and date of death, if applicable.

Haloperidol, olanzapine, and quetiapine accounted for >97% of antipsychotics utilized, and thus we restricted antipsychotic data to these medications. Given confounding use of clonidine for hemodynamic management, we selected guanfacine as the oral alpha-2 agonist of interest. We obtained daily administration of antipsychotic medications, guanfacine, and valproic acid for each patient while in the ICU. Only medications actually administered were used in analysis. Additional medication administration data included receipt (yes/no) and dose on days exposed of the following infusions: propofol, dexmedetomidine, benzodiazepine (in midazolam equivalents), and fentanyl. We obtained discharge medication lists for all hospital survivors. Patients with missing data were excluded from analysis.

Outcomes

Outcomes of interest included daily mental status (normal, delirious, and comatose), in-hospital mortality, days alive and free of the ICU (ie, ICU-free days) over the first 30 days following ICU admission, and days alive and free of the hospital (ie, hospital-free days) during the first 30 days after ICU admission for each ICU encounter. We defined normal mental status for a given day as no positive delirium or coma assessments. We defined delirium for a given day as RASS score −3 to +4 with positive CAM-ICU assessment. We defined coma on a given day as RASS score −4 to −5. ICU-free and hospital-free days were chosen as opposed to length-of-stay measurements to reduce bias introduced by death.

Statistical Analysis

In the event of repeat ICU encounters, we described patient characteristics only from their initial ICU visit. We described patient characteristics and delirium-prescribing practices using median values and interquartile ranges (IQRs) for continuous variables and frequency (%) for categorical variables. Unadjusted comparisons between pharmacologic recipients and nonrecipients were made using the Pearson χ2 test or Wilcoxon rank sum test, for categorical and quantitative variables, respectively.

We performed a daily state transition model using multinomial logistic regression to evaluate the independent association of haloperidol, olanzapine, quetiapine, guanfacine, and valproic acid on a given day, with mental status (delirium, coma, and normal) the following day—adjusting for mental status the preceding day, baseline demographic covariates and markers of severity of illness, including age, sex, UHC-expected mortality, shock (yes/no), ICU type (medical, surgical, trauma, and cardiovascular), delirium (yes/no), coma (yes/no), mechanical ventilation (yes/no), benzodiazepine infusion (yes/no and total dose on previous day, in midazolam equivalents), propofol infusion (yes/no and total dose on previous day), dexmedetomidine infusion (yes/no and total dose on previous day), and fentanyl infusion (yes/no and total dose on previous day). Daily mental status was used as opposed to daily delirium status (yes/no) to capture those patients with RASS <−3, or comatose, in whom delirium could not be assessed and may falsely be considered “normal” due to being “not delirious.” Transitions to an absorbing state (ie, death and discharge) were not modeled. For each drug exposure, we report the associated odds ratio (OR) for transitioning to delirium (versus normal mental status). While the ORs for transitioning to coma were also estimated, these are not the focus of this study and are not reported here.

To account for the correlated nature of repeated (daily) assessments of cognitive status, the cluster bootstrap method was used to construct 95% confidence intervals (95% CIs) for the medication effects on the odds of delirium the following day, and statistical significance was determined by inclusion or exclusion of the null value (OR = 1). In the cluster bootstrap, clusters (here, individual patients) are sampled with replacement, rather than sampling single measurements.22 All measurements for a patient are sampled together, and the number of clusters (including repeats for some patients) equals the original sample size. This ensures that within-cluster correlation among measurements is preserved in the bootstrap samples. A total of 10,000 cluster bootstrap samples were generated using this method.

Each sample was then analyzed using the methods described previously. The resulting estimates were then used to generate 95% CIs using the bootstrap bias-corrected and accelerated (BCa) method and bootstrap P values, which are both based on percentiles of the estimates.23 While a mixed-effects method might have been used as an alternative method of accounting for repeated measures, there is no mature software implementations of mixed-effects multinomial logistic regression in R.

A Cox proportional hazards model with time-varying exposure was performed to assess the independent association of haloperidol, olanzapine, quetiapine, guanfacine, and valproic acid administration on in-hospital mortality using the same prespecified covariates. Linear regression was used to assess the independent associations of haloperidol, olanzapine, quetiapine, guanfacine, and valproic acid administration on ICU-free and hospital-free days during the first 30 days after ICU admission adjusting for age, sex, UHC-expected mortality, shock diagnosis (yes/no), ICU type (medical, surgical, trauma, and cardiovascular), mechanical ventilation duration, benzodiazepine infusion (yes/no and mean daily dose on days exposed), propofol infusion (yes/no and mean daily dose on days exposed), dexmedetomidine infusion (yes/no and mean daily dose on days exposed), and fentanyl infusion (yes/no and mean daily dose on days exposed).

For the Cox and linear regression analyses, statistical significance was assessed using Wald-type tests. For linear regression analyses, residual diagnostics (ie, scatter plots and normal Q-Q plots) were examined for evidence of heteroscedasticity or deviation from residual normality. We considered P values < .05 as statistically significant in these 2 types of analysis. No familywise hypotheses were considered, and no adjustment made to control a familywise type-I error probability. This ensures a type-I error rate of 5% for each test, although the probability of 1 or more type-I errors among these tests is >5% (eg, if there are 10 tests the probability of making 1 or more type-I errors is as high as 40%, but likely smaller due to correlation among the test statistics). This approach is consistent with the study design and aims to independently test for significant associations between the use of certain drugs and patient outcomes.24,25

Due to the descriptive and exploratory nature of this study, no statistical power or precision analysis was implemented a priori. To gauge the degree of statistical precision achieved in this study, we refer the reader to examine the 95% CIs provided. CIs are considered sufficiently precise, and provide evidence of sample size adequacy, when they include only those values that are consistent with the study conclusions.26 When a CI is very wide, there may be insufficient evidence to draw a firm conclusion. We used R version 3.6.1 for all statistical analyses.

RESULTS

Table 1. - Patient Characteristics Displayed Based on Receipt of Pharmacologic Treatment for Delirium
Characteristic Recipients (n = 3770) Nonrecipients (n = 4109) Total cohort (n = 7879) P value
Age (y) 62 [49–73] 61 [48–71] 61 [48–72] .003 a
Sex (male) 2375 (63.0%) 2285 (55.6%) 4660 (59.4%) <.001 b
BMI (kg/m2) 27 [23–32] 27 [23–33] 27 [23–33] .600a
Home medications
 Antidepressants 775 (20.6%) 816 (19.9%) 1591 (20.2%) .440b
 Benzodiazepines 738 (19.6%) 761 (18.5%) 1499 (19.0%) .230b
 Opioids 2124 (56.3%) 2506 (61.0%) 4630 (58.8%) <.001 b
ICU type <.001 b
 Cardiovascular 801 (21.2%) 452 (11.0%) 1253 (15.9%)
 Medical 1219 (32.3%) 2058 (50.1%) 3277 (41.6%)
 Surgical 674 (17.9%) 780 (19.0%) 1454 (18.4%)
 Trauma 1058 (28.1%) 805 (19.6%) 1863 (23.6%)
UHC-expected mortality 4.3% [1.2%–14%] 3.7% [0.8%–16%] 4% [1%–15%] <.001 a
Shock 1114 (29.5%) 783 (19.1%) 1906 (24.2%) <.001 b
Mechanical ventilation 2978 (79.0%) 3298 (80.3%) 6276 (80.0%) .200b
Propofol infusion exposure 2226 (59.0%) 1592 (38.7%) 6908 (77.8%) <.001 b
Dexmedetomidine infusion exposure 983 (26.1%) 269 (6.5%) 1948 (22.0%) <.001 b
Fentanyl infusion exposure 1951 (51.8%) 1383 (33.6%) 5476 (61.7%) <.001 b
Benzodiazepine infusion exposure 252 (6.7%) 148 (3.6%) 670 (8.5%) <.001 b
Duration of delirium (d) 2.1 [0.8–5.4] 0.5 [0.2–1.4] 1 [0.4–3] <.001 a
ICU length of stay (d) 4.2 [2–8.2] 2.5 [1.3–4.2] 3.1 [1.6–6.0] <.001 a
Hospital length of stay (d) 11.1 [6.6–19.0] 6.5 [3.7–10.6] 8.2 [4.7–14.6] <.001 a
In-hospital mortality 330 (8.8%) 487 (11.9%) 817 (10.4%) <.001 b
“Nonrecipients” include patients who did not receive pharmacologic treatment for ICU delirium. “Recipients” include those patients who did receive pharmacologic treatment while in the ICU. Variables are presented as median [interquartile range] or count (%), as appropriate. P value is the statistical difference between recipient and nonrecipient groups. Bold indicates statistically significant P value.
Abbreviations: BMI, body mass index; ICU, intensive care unit; UHC-expected mortality = University Health Consortium–expected probability of mortality.27
aWilcoxon test used.
bPearson test used.

F1
Figure 1.:
Prescribing patterns for management of ICU delirium. Illustration of prescribing practices within our cohort. Of the 24,884 ICU admissions, 9682 were diagnosed with delirium as defined by a positive assessment on the CAM-ICU. All excluded patients were excluded for home antipsychotic use as obtained from the electronic health record to obtain the final study sample cohort of 8591. Of delirious patients who received pharmacologic treatment, 3898 received antipsychotic medications, 187 received guanfacine, and 72 received valproic acid. Combinations of antipsychotic medications were also frequent with 1192 (30.6%) of antipsychotic recipients receiving a combination of 2 antipsychotic medications and 664 (17.0%) receiving 3 or more different antipsychotic medications. CAM-ICU indicates Confusion Assessment Method for the ICU; ICU, intensive care unit.

A total of 8591 unique ICU admissions with delirium were identified from 7879 patients throughout the study period (Figure 1). Baseline patient characteristics for the total cohort, along with recipients and nonrecipients of pharmacologic treatment for delirium, are described in Table 1. The median age of our cohort was 61 years (48–72); majority were men (59.4%) and required mechanical ventilation (80.0%). Patients were most frequently admitted to the medical ICU (41.6%) as compared to the other ICUs. There were statistically significant differences between pharmacologic recipients and nonrecipients across most characteristics. The median duration of ICU delirium was 1 (0.4–3) days for the total cohort, 2.1 (0.8–5.4) for recipients, and 0.5 (0.2–1.4) for nonrecipients.

Prescribing Patterns

Table 2. - Pharmacologic Interventions and Prescribing Patterns for ICU Delirium
Pharmacologic intervention N = 8591
ICU encounters
Antipsychotic medication use 3898 (45.3%)
 Delirium day initiated 2 [1–4]
 Median daily dose of haloperidol on days exposed (mg) 5 [3–10]
 Median daily dose of olanzapine on days exposed (mg) 7.5 [5–12.5]
 Median daily dose of quetiapine on days exposed (mg) 50 [25–125]
 Duration of any antipsychotic (d) 4 [2–10]
 Continued at hospital discharge 804 (20.6%)a
Guanfacine use 187 (2.2%)
 Delirium day initiated 5 [2–11]
 Median daily dose of guanfacine on days exposed (mg) 2 [1–4]
 Duration of guanfacine (d) 3 [2–7]
 Continued at hospital discharge 28 (15.0%)a
Valproic acid use 72 (0.84%)
 Delirium day initiated 1 [1–4]
 Median daily dose of valproic acid on days exposed (mg) 750 [500–1500]
 Duration of valproic acid (d) 3 [1–4]
 Continued at hospital discharge 38 (52.8%)a
Variables are presented as median [interquartile range] or count (%), as appropriate. Data shown represent the results from 8591 ICU encounters identified within the study period obtained from 7879 unique patients.
Abbreviation: ICU, intensive care unit.
aThe percentage of patients continuing a medication at hospital discharge is relative to the count of those who received the drug during their ICU stay.

Overall, 45.6% of patients received pharmacologic treatment for delirium. Pharmacologic recipients had a median highest RASS score of 1 (0–1) on the day of medication initiation (suggesting hyperactive symptoms). For nonrecipients, their median highest RASS score on delirium day 0 was 0 (−1, 0). Approximately half of all delirious ICU patients were treated with an antipsychotic medication (45.4%). Haloperidol, olanzapine, and quetiapine accounted for 97.6% of all antipsychotics utilized within our cohort. As shown in Figure 1, of those who received an antipsychotic, 61.4% received haloperidol, 52.6% received olanzapine, and 47.8% received quetiapine. A total of 1192 (30.6%) of patients who received an antipsychotic medication while in the ICU received a combination of 2 antipsychotic medications, and 664 (17.1%) received 3 or more different antipsychotic medications throughout their ICU stay. Additional description of prescribing practices by admitting to ICU and mechanical ventilation status on day delirium began (delirium day 0) are included within Supplemental Digital Content (Figures S1–S6, https://links.lww.com/AA/D315). Antipsychotics were initiated on median delirium day 2 (1–4) and continued for a median of 4 (2–10) days (Table 2). More than 1 in 5 patients newly initiated on an antipsychotic medication in the ICU were discharged from the hospital on the medication (20.6%). Approximately 2.2% of all delirious patients received guanfacine and 0.84% received valproic acid. Guanfacine was started on median delirium day 5 while valproic acid was started on median delirium day 1. Both agents were continued for median duration of 3 days among those exposed (Table 2). Guanfacine, when used in the ICU, was continued in 15.0% of patients at hospital discharge whereas valproic acid was continued in 52.8% of recipients (Table 2).

Prescribing Choice and Outcomes

Results from daily transition models testing the independent association of medication exposure on delirium the following day, after accounting for the previous day’s mental status and other covariates, are shown in Table 3 and Figure 2. Haloperidol and olanzapine were both independently associated with an increased odds of delirium the following day (ie, lower likelihood of delirium resolution) after adjusting for prespecified covariates (OR, 1.48; 95% CI, 1.30-1.65; P < .001 and OR, 1.37; 95% CI, 1.20-1.56; P = .003, respectively). Quetiapine, valproic acid, and guanfacine were not significantly associated with delirium the following day (OR, 1.06; 95% CI, 0.97-1.17; P = .74; OR, 1.26; 95% CI, 0.74-2.15; P = .88; OR, 1.12; 95% CI, 0.72-1.70; P = .89, respectively).

Table 3. - Associations Between Pharmacologic Choice and In-Hospital Patient Outcomes
Delirium Mortality ICU-free days Hospital-free days
Medication Odds ratio (95% CI) Hazard ratio (95% CI) Difference in mean (95% CI) Difference in mean (95% CI)
Haloperidol 1.48a (1.30-1.65) 1.46a (1.10-1.93) −0.33 (−0.81 to 0.15) −1.00a (−1.51 to −0.49)
Olanzapine 1.37a (1.20-1.56) 1.67a (1.14-2.45) 0.38 (−0.29 to 1.05) −1.55a (−2.26 to −0.84)
Quetiapine 1.06 (0.97-1.17) 0.58a (0.40-0.84) 0.04 (−0.50 to 0.57) −1.22a (−1.79 to −0.65)
Guanfacine 1.12 (0.72-1.70) 0.61 (0.22-1.67) −0.31 (−2.61 to 1.99) −0.62 (−3.04 to −1.81)
Valproic acid 1.26 (0.74-2.15) 2.50 (0.24-25.46) −1.52 (−3.95 to 0.91) −1.16 (−4.23 to 0.91)
Delirium was calculated from daily state transition model using multinomial logistic regression to evaluate the independent association of pharmacologic choice (haloperidol [yes/no], olanzapine [yes/no], quetiapine [yes/no], guanfacine [yes/no], and valproic acid [yes/no]) on continued delirium the following day, after adjusting for key confounders. No medication reduced the risk of delirium the following day. Conversely, haloperidol and olanzapine were independently associated with increased odds of continued delirium (ie, worse delirium resolution). “Mortality” refers to in-hospital mortality. We assessed the association between mortality and the medications of interest using a single Cox proportional hazards regression model with time-varying exposure for each medication. Haloperidol and olanzapine were independently associated with increased hazard of in-hospital mortality, whereas quetiapine was associated with decreased hazard. ICU-free and hospital-free days during the first 30 d after ICU admission were calculated using linear regression to assess for the independent association of haloperidol, olanzapine, quetiapine, guanfacine, and valproic acid on these outcomes. The difference in mean describes the estimated difference in mean number of days, or effect size, and denotes the adjusted difference in the average number of days alive and free of the ICU or hospital during the 30-d period studied. Haloperidol, olanzapine, and quetiapine were independently associated with decreased hospital-free days (ie, fewer days alive and free of the hospital).
Abbreviations: CI, confidence interval; ICU, intensive care unit.
aStatistically significant result of P ≤ .05.

F2
Figure 2.:
Associations of pharmacologic treatment choice on delirium continuation and mortality the following day. Each figure is the result of a single model performed, after adjusting for described confounders. A, Delirium continuation assessed using daily transition model; odds ratio of transition to delirium by medications received the day prior. Haloperidol and olanzapine were both associated with increased odds of delirium the following day (OR, 1.48; 95% CI, 1.30-1.65 and OR, 1.37; 95% CI, 1.20-1.56, respectively). Quetiapine, valproic acid, and guanfacine were not associated with continued delirium (OR, 1.06; 95% CI, 0.97-1.17; OR, 1.26; 95% CI, 0.74-2.15; and OR, 1.12; 95% CI, 0.72-1.70, respectively). B, In-hospital mortality; hazard of in-hospital mortality by medication received during ICU stay. Haloperidol and olanzapine were associated with an increased hazard of in-hospital mortality (HR, 1.46; 95% CI, 1.10-1.93 and HR, 1.67; 95% CI, 1.14-2.45, respectively) while quetiapine was associated with a decreased hazard of mortality (HR, 0.58; 95% CI, 0.40-0.84). Guanfacine and valproic acid were not significantly associated with in-hospital mortality. CI indicates confidence interval; HR, hazard ratio; ICU, intensive care unit; OR, odds ratio.

Results from Cox regression models testing the independent association of medication exposure on a given day with in-hospital mortality are shown in Table 3 and Figure 2. Haloperidol and olanzapine use were independently associated with an increased hazard of mortality (hazard ratio [HR], 1.46; 95% CI, 1.10-1.93; P = .01; HR, 1.67; 95% CI, 1.14-2.45; P = .01, respectively), while quetiapine use was associated with a decreased hazard of mortality (HR, 0.58; 95% CI, 0.40-0.84; P = .01). Guanfacine and valproic acid were not independently associated with in-hospital mortality (P = .33 and P = .44, respectively).

We found no significant association with any of the medications tested on ICU-free days (Table 3 and Figure 3). However, haloperidol (difference in mean, −1.00 days; 95% CI, −1.51 to −0.49; P = .001), olanzapine (difference in mean, −1.55 days; 95% CI, −2.26 to −0.84; P < .001), and quetiapine (difference in mean −1.22 days; 95% CI, −1.79 to −0.65; P < .001) were independently associated with fewer hospital-free days. Neither guanfacine nor valproic acid had an independent association with hospital-free days (P = .70 and P = .21, respectively).

F3
Figure 3.:
Association of pharmacologic management on ICU-free and hospital-free days. Each figure is the result of a single model performed, after adjusting for described confounders. A, ICU-free days during the first 30 d after ICU admission using multivariable linear regression. There was no association of any medication administration on ICU-free days. B, Hospital-free days during the first 30 d after ICU admission using multivariable linear regression. Antipsychotic medications haloperidol (difference in mean, −1.00 d; 95% CI, −1.51 to −0.49), olanzapine (difference in mean, −1.55 d; 95% CI, −2.26 to −0.84), and quetiapine (effect size, −1.22 d; 95% CI, −1.79 to −0.65) were all associated with fewer hospital-free days. Guanfacine and valproic acid had no association with hospital-free days. CI indicates confidence interval; ICU, intensive care unit.

Subanalyses of patient groups by admitting ICU and mechanical ventilation status on delirium day 0 are included within Supplemental Digital Content, Figures S7–S15, https://links.lww.com/AA/D315. Subanalyses should be interpreted with caution and should be used only for qualitative interpretation by the reader.

DISCUSSION

Our study examining clinical prescribing practices for the management of ICU delirium found antipsychotics were commonly prescribed for ICU delirium, especially when patients demonstrated hyperactive/agitated symptoms. Guanfacine and valproic acid were much less frequently utilized. The median duration of medication exposure was more than double the median duration of delirium, and more than 1 in 5 patients initiated on antipsychotic medications were continued on them on hospital discharge. Haloperidol and olanzapine were both associated with lower likelihood of delirium resolution and with increased in-hospital mortality. Finally, the 3 antipsychotics investigated (haloperidol, olanzapine, and quetiapine) were all independently associated with fewer days alive and free of hospitalization.

Our study is novel in that it was conducted over a period where antipsychotic medications, guanfacine, and valproic acid were emerging as innovative and effective treatment methods for decreasing delirium duration and/or severity.7,9,10 Variable prescribing practices found could represent differences in provider preference, perceived patient benefit, lack of recognition of potential harm, and limited available data to guide clinicians. Emerging literature indicates that newly initiated antipsychotic medications are often continued beyond hospital discharge,15,27–30 the majority with no documented indication.29,31,32 These medications are not benign, affect multiple organ systems, interact with other medications, and can pose risk to patients that may affect long-term patient outcomes.33,34

In our cohort, haloperidol and olanzapine were associated with a lower likelihood of delirium resolution, or greater odds of continued delirium (48% and 37% higher, respectively), while other pharmacologic treatment choices did not significantly affect delirium resolution. This aligns with the literature demonstrating haloperidol to be ineffective for treating ICU delirium11,12 and expounds on limited data available for other agents.13 In the Modifying the Impact of ICU-Associated Neurological Dysfunction-USA (MIND-USA)11 study, the largest trial to date investigating antipsychotic medications for treatment of ICU delirium, neither typical (haloperidol) nor atypical (ziprasidone) antipsychotic medications showed any effect on delirium duration compared to placebo. This trial, however, was underpowered to examine effects on patients with hyperactive delirium specifically. In our study of largely hyperactive delirious patients, none of the medications examined were associated with improved delirium resolution, with haloperidol and olanzapine actually associated with worse resolution. Due to the limited number of patients who received guanfacine and valproic acid, there was insufficient evidence that their use affected delirium resolution.

In addition to limited benefits demonstrated with these medications for treatment of delirium, our study suggests potential harms—specifically increased risk of in-hospital mortality with haloperidol and olanzapine use. While previous studies with haloperidol for treatment of ICU delirium have shown potential harm (eg, extrapyramidal symptoms, oversedation, increased need for respiratory or circulatory support),9,12,35,36 they have not demonstrated an increased risk of mortality.35,36 It is difficult to determine why these medications were associated with increased in-hospital mortality within our retrospective cohort (eg, patient selection, attributable risk of delirium in mortality, or severity of illness), though we adjusted for many of these factors in our multivariable analysis. Lack of data supporting their efficacy in treating delirium, together with findings describing worsened in-hospital outcomes, suggest they should be used cautiously and for the shortest allowable duration when required. Our findings showing antipsychotics associated with fewer days alive and free of the hospital echo prior retrospective investigations demonstrating antipsychotic medication use associated with longer hospitalizations.37,38 Large prospective, randomized trials, however, have failed to demonstrate this association and it should be noted that length of hospitalization and hospital-free days are not equivalent measurements due to potential effects of death within the sample.11,12,36 It is interesting that all studies exhibiting an association between antipsychotic medications and longer hospitalizations were nonprotocolized, wherein medication initiation and duration were performed at providers’ discretion. The use of protocols with targeted discontinuation regimens, such as those used in prospective trials, might have a role in decreasing this effect—potentially emphasizing the importance of prompt discontinuation.

Our study is strengthened by a large heterogeneous sample of patients across multiple ICUs. Clinicians were not protocolized, medication regimens were reflective of clinical practice and not a research trial, and we only recorded medications actually received. All patients were treated within ICUs whose care centers on the ABCDEF bundle emphasizing multicomponent, nonpharmacologic strategies aimed at decreasing delirium such as reorientation, mobilization, sleep hygiene, and removal of restraints. Delirium was assessed at least twice per shift by bedside nursing with a record of high sensitivity and specificity for identifying delirium.17 The 3 most frequently prescribed antipsychotics at our institution—haloperidol, olanzapine, and quetiapine—were analyzed in addition to guanfacine and valproic acid. Daily transition models assessing for the independent association of these medications were applied with high power due to our large sample size and adjusted for multiple covariates including severity of illness, sedative exposure, and daily mental status strengthening our findings. Finally, this was the first investigation to our knowledge to describe the use of guanfacine for management of ICU delirium. Our study also has several limitations. Hospital EHR information is limited by missing or incorrect data entry. We included only delirious patients, excluded patients on home antipsychotics, and thus inferred that newly prescribed medications were for the treatment of ICU delirium, but it is possible that some were initiated for other on- or off-label use. For this reason, we chose not to analyze clonidine given its overlapping applications. Although use and dose of dexmedetomidine were adjusted for in our analyses, we did not examine this medication specifically on outcomes given its primary use as a sedative. The small proportion of patients who received guanfacine and valproic acid potentially underpowered our ability to demonstrate significant associations with these medications, and we did not screen for home use of valproic acid. Our study was conducted at a single academic institution, and thus our sample may not be representative of other centers; however, our large cohort across multiple ICUs includes a heterogeneous mix of data. Our institution adheres to the ABCDEF bundle for delirium management, which has been shown to improve outcomes including in-hospital mortality, coma, and delirium duration in a dose-dependent manner according to bundle adherence.5 We were unable to monitor specific bundle component adherence within our retrospective cohort or account for this confounder within our models, which may limit our results, but previous prospective studies have demonstrated high compliance (>85%) for all portions of the bundle at our medical center. The UHC-expected mortality index served as a marker for severity of illness and is well validated among hospitalized patients. This model, however, relies on administrative data and is limited by the accuracy of documentation. Additionally, among ICU patients, the UHC mortality index may have decreasing accuracy with increased severity of illness, underpredicting mortality.39 Because we present and statistically test multiple associations, we caution the reader to consider the risk of spurious significant results (ie, type-I errors). Whether as part of a single study, or across many studies, repeated statistical testing accumulates the risk of 1 or more spurious significant results. Finally, this study was retrospective and observational in design and unable to test causation.

In conclusion, we found pharmacologic interventions for the treatment of ICU delirium are common, most often with antipsychotic medications, and often continued after delirium resolution and hospital discharge. Haloperidol and olanzapine were associated with lower likelihood of delirium resolution and increased in-hospital mortality. All antipsychotic medications investigated were associated with fewer days alive and free of hospitalization. Further investigation into the safety and efficacy of these medications, particularly for those continued following hospital discharge, across care systems, and within specific ICU populations is important for understanding their role in patient recovery.

DISCLOSURES

Name: Christina S. Boncyk, MD.

Contribution: This author helped with conception of the study, data collection, analysis, and interpretation; manuscript drafting, revision, and submission; and approved the final version of the manuscript.

Name: Emily Farrin, MD.

Contribution: This author helped with conception of the study, manuscript drafting and revision, and approved the final version of the manuscript.

Name: Joanna L. Stollings, PharmD.

Contribution: This author helped in data collection, manuscript revision, and approved the final version of the manuscript.

Name: Kelli Rumbaugh, PharmD.

Contribution: This author helped in data collection, manuscript revision, and approved the final version of the manuscript.

Name: Jo Ellen Wilson, MD, MPH.

Contribution: This author helped in manuscript revision and approved the final version of the manuscript.

Name: Matt Marshall, PharmD.

Contribution: This author helped in data collection, manuscript revision, and approved the final version of the manuscript.

Name: Xiaoke Feng, MS.

Contribution: This author helped in data analysis and interpretation, manuscript revision, and approved the final version of the manuscript.

Name: Matthew S. Shotwell, PhD.

Contribution: This author helped in data analysis and interpretation, manuscript revision, and approved the final version of the manuscript.

Name: Pratik P. Pandharipande, MD, MSCI.

Contribution: This author helped with conception of the study, data analysis and interpretation, manuscript revision, and approved the final version of the manuscript.

Name: Christopher G. Hughes, MD, MS.

Contribution: This author helped with conception of study, contribution to data collection, data analysis and interpretation, drafting of manuscript, manuscript revisions, and approved the final version of the manuscript.

This manuscript was handled by: Avery Tung, MD, FCCM.

    REFERENCES

    1. Association AP. Diagnostic and Statistical Manual of Mental Disorders. 2013.5th ed. American Psychiatric Association;
    2. Ouimet S, Kavanagh BP, Gottfried SB, Skrobik Y. Incidence, risk factors and consequences of ICU delirium. Intensive Care Med. 2007;33:66–73.
    3. Pandharipande PP, Girard TD, Jackson JC, et al.; BRAIN-ICU Study Investigators. Long-term cognitive impairment after critical illness. N Engl J Med. 2013;369:1306–1316.
    4. Ely EW, Shintani A, Truman B, et al. Delirium as a predictor of mortality in mechanically ventilated patients in the intensive care unit. JAMA. 2004;291:1753–1762.
    5. Pun BT, Balas MC, Barnes-Daly MA, et al. Caring for critically ill patients with the ABCDEF bundle: results of the ICU liberation collaborative in over 15,000 adults. Crit Care Med. 2019;47:3–14.
    6. Hipp DM, Ely EW. Pharmacological and nonpharmacological management of delirium in critically ill patients. Neurotherapeutics. 2012;9:158–175.
    7. Devlin JW, Roberts RJ, Fong JJ, et al. Efficacy and safety of quetiapine in critically ill patients with delirium: a prospective, multicenter, randomized, double-blind, placebo-controlled pilot study. Crit Care Med. 2010;38:419–427.
    8. Sher Y, Miller Cramer AC, Ament A, Lolak S, Maldonado JR. Valproic acid for treatment of hyperactive or mixed delirium: rationale and literature review. Psychosomatics. 2015;56:615–625.
    9. Skrobik YK, Bergeron N, Dumont M, Gottfried SB. Olanzapine vs haloperidol: treating delirium in a critical care setting. Intensive Care Med. 2004;30:444–449.
    10. Tahir TA, Eeles E, Karapareddy V, et al. A randomized controlled trial of quetiapine versus placebo in the treatment of delirium. J Psychosom Res. 2010;69:485–490.
    11. Girard TD, Exline MC, Carson SS, et al.; MIND-USA Investigators. Haloperidol and ziprasidone for treatment of delirium in critical illness. N Engl J Med. 2018;379:2506–2516.
    12. Page VJ, Ely EW, Gates S, et al. Effect of intravenous haloperidol on the duration of delirium and coma in critically ill patients (Hope-ICU): a randomised, double-blind, placebo-controlled trial. Lancet Respir Med. 2013;1:515–523.
    13. Burry L, Hutton B, Williamson DR, et al. Pharmacological interventions for the treatment of delirium in critically ill adults. Cochrane Database Syst Rev. 2019;9:CD011749.
    14. Devlin JW, Skrobik Y, Gélinas C, et al. Clinical practice guidelines for the prevention and management of pain, agitation/sedation, delirium, immobility, and sleep disruption in adult patients in the ICU. Crit Care Med. 2018;46:e825–e873.
    15. Tomichek JE, Stollings JL, Pandharipande PP, Chandrasekhar R, Ely EW, Girard TD. Antipsychotic prescribing patterns during and after critical illness: a prospective cohort study. Crit Care. 2016;20:378.
    16. Ely EW, Margolin R, Francis J, et al. Evaluation of delirium in critically ill patients: validation of the Confusion Assessment Method for the Intensive Care Unit (CAM-ICU). Crit Care Med. 2001;29:1370–1379.
    17. Vasilevskis EE, Morandi A, Boehm L, et al. Delirium and sedation recognition using validated instruments: reliability of bedside intensive care unit nursing assessments from 2007 to 2010. J Am Geriatr Soc. 2011;59(suppl 2):S249255.
    18. Marra A, Ely EW, Pandharipande PP, Patel MB. The ABCDEF bundle in critical care. Crit Care Clin. 2017;33:225–243.
    19. Lipshutz AK, Feiner JR, Grimes B, Gropper MA. Predicting mortality in the intensive care unit: a comparison of the university health consortium expected probability of mortality and the mortality prediction model III. J Intensive Care. 2016;4:35.
    20. Barbieri JS, Fuchs BD, Fishman N, et al. The mortality review committee: a novel and scalable approach to reducing inpatient mortality. Jt Comm J Qual Patient Saf. 2013;39:387–395.
    21. Brooks SE, Mukherjee K, Gunter OL, et al. Do models incorporating comorbidities outperform those incorporating vital signs and injury pattern for predicting mortality in geriatric trauma? J Am Coll Surg. 2014;219:1020–1027.
    22. Van der Leeden R, Meijer E, Busing FM. Deleeuw J, Meijer E, eds. Resampling multilevel models. In: Handbook of Multilevel Analysis. 2008:Springer; 401–433.
    23. Davison AC, Hinkley DV. Bootstrap Methods and Their Application. 1997.Cambridge University Press;
    24. Althouse AD. Adjust for multiple comparisons? It’s not that simple. Ann Thorac Surg. 2016;101:1644–1645.
    25. Feise RJ. Do multiple outcome measures require p-value adjustment? BMC Med Res Methodol. 2002;2:8.
    26. Hoenig JM, Heisey DM. The abuse of power: the pervasive fallacy of power calculations for data analysis. Am Stat. 2001:19–24.
    27. Morandi A, Vasilevskis E, Pandharipande PP, et al. Inappropriate medication prescriptions in elderly adults surviving an intensive care unit hospitalization. J Am Geriatr Soc. 2013;61:1128–1134.
    28. Flurie RW, Gonzales JP, Tata AL, Millstein LS, Gulati M. Hospital delirium treatment: continuation of antipsychotic therapy from the intensive care unit to discharge. Am J Health Syst Pharm. 2015;72:S133–S139.
    29. Marshall JH, Herzig SJ, Howell MD. Antipsychotic utilization in the intensive care unit and in transitions of care. J Crit Care. 2016;33:119–124.
    30. Levine AR, Lemieux SM, D’Aquino D, Tenney A, Pisani M, Ali S. Risk factors for continuation of atypical antipsychotics at hospital discharge in two intensive care unit cohorts [published online ahead of print July 16, 2019]. Clin Med Insights Psychiatry. 2019. doi:10.1177/1179557319863813.
    31. Kram BL, Kram SJ, Brooks KR. Implications of atypical antipsychotic prescribing in the intensive care unit. J Crit Care. 2015;30:814–818.
    32. Rowe AS, Hamilton LA, Curtis RA, et al. Risk factors for discharge on a new antipsychotic medication after admission to an intensive care unit. J Crit Care. 2015;30:1283–1286.
    33. Schneider-Thoma J, Efthimiou O, Huhn M, et al. Second-generation antipsychotic drugs and short-term mortality: a systematic review and meta-analysis of placebo-controlled randomised controlled trials. Lancet Psychiatry. 2018;5:653–663.
    34. Szmulewicz AG, Angriman F, Pedroso FE, Vazquez C, Martino DJ. Long-term antipsychotic use and major cardiovascular events: a retrospective cohort study. J Clin Psychiatry. 2017;78:e905–e912.
    35. Collet MO, Caballero J, Sonneville R, et al.; AID-ICU cohort study co-authors. Prevalence and risk factors related to haloperidol use for delirium in adult intensive care patients: the multinational AID-ICU inception cohort study. Intensive Care Med. 2018;44:1081–1089.
    36. van den Boogaard M, Slooter AJC, Brüggemann RJM, et al.; REDUCE Study Investigators. Effect of haloperidol on survival among critically ill adults with a high risk of delirium: the REDUCE randomized clinical trial. JAMA. 2018;319:680–690.
    37. Daniels LM, Nelson SB, Frank RD, Park JG. Pharmacologic treatment of intensive care unit delirium and the impact on duration of delirium, length of intensive care unit stay, length of hospitalization, and 28-day mortality. Mayo Clin Proc. 2018;93:1739–1748.
    38. Weaver CB, Kane-Gill SL, Gunn SR, Kirisci L, Smithburger PL. A retrospective analysis of the effectiveness of antipsychotics in the treatment of ICU delirium. J Crit Care. 2017;41:234–239.
    39. Enfield KB, Schafer K, Zlupko M, et al. A comparison of administrative and physiologic predictive models in determining risk adjusted mortality rates in critically ill patients. PLoS One. 2012;7:e32286.

    Supplemental Digital Content

    Copyright © 2021 International Anesthesia Research Society