Trial Design and Population
This randomized, double-blind, placebo-controlled, phase 3 trial was conducted at 16 medical centers in the United States. Before we randomly assigned the patients to receive a trial drug or placebo, we obtained written informed consent from the patients or their authorized representatives. The institutional review board at each participating center approved the protocol, available with the full text of this article at NEJM.org. The Food and Drug Administration approved an Investigational New Drug application, which was obtained because the intravenous route of administration and the indication for delirium that were used in this trial are not approved for antipsychotic medications (see the Supplementary Appendix, available at NEJM.org). The trial was registered at ClinicalTrials.gov on September 29, 2010, before the first patient was enrolled. The statistical analysis plan was registered at Open Science Framework (https://osf.io/mq38r) on March 22, 2018, before the trial-group assignments were unmasked. The authors vouch for the accuracy and completeness of the data and for the fidelity of the trial to the protocol.
Patients who had been admitted to the participating hospitals were eligible for inclusion if they were 18 years of age or older and were receiving treatment in a medical or surgical ICU with invasive or noninvasive positive pressure ventilation, vasopressors, or an intraaortic balloon pump, and they were eligible for random assignment to a trial group if they had delirium. We excluded patients who, at baseline, had severe cognitive impairment, as determined by medical record review and the short form of the Informant Questionnaire on Cognitive Decline in the Elderly (IQCODE; scores range from 1.0 to 5.0, with higher scores indicating more severe cognitive impairment [a score of ≥4.5 resulted in exclusion because of severe dementia])22; were at high risk for medication side effects because of pregnancy, breast-feeding, a history of torsades de pointes, QT prolongation, a history of neuroleptic malignant syndrome, or allergy to haloperidol or ziprasidone; were receiving ongoing treatment with an antipsychotic medication; were in a moribund state; had rapidly resolving organ failure; were blind, deaf, or unable to speak or understand English; were incarcerated; or were enrolled in another study or trial that prohibited coenrollment. Details of the inclusion and exclusion criteria are provided in the Supplementary Appendix. Noncomatose patients were excluded if informed consent could not be obtained within 72 hours after inclusion criteria had been met, and comatose patients were excluded if informed consent could not be obtained within 120 hours after inclusion criteria had been met.
To minimize the time between the onset of delirium and randomization, we obtained informed consent, when possible, before the onset of delirium; delirium was detected with the use of the Confusion Assessment Method for the ICU (CAM-ICU),23,24 a validated tool that identifies delirium on the basis of an acute change or fluctuating course of mental status plus inattention and either altered level of consciousness or disorganized thinking. If delirium was not present at the time that informed consent was obtained, trained research personnel evaluated patients twice daily until delirium was present or until death, discharge from the ICU, development of an exclusion criterion, or a maximum of 5 days.
When delirium was present at the time of informed consent or during the 5 days after informed consent was obtained and the corrected QT interval was less than 550 msec on a 12-lead electrocardiogram (see the Supplementary Appendix), we randomly assigned the patients, in a 1:1:1 ratio, to receive placebo, haloperidol, or ziprasidone using a computer-generated, permuted-block randomization scheme, with stratification according to trial site. The research personnel, managing clinicians, patients, and their families were not aware of the trial-group assignments.
The trial drugs or placebo were administered intravenously with the use of colorless preparations delivered in identical bags. Immediately after the trial-group assignment, the first dose of a trial drug or placebo was administered: patients younger than 70 years of age received 0.5 ml of placebo (0.9% saline) or 2.5 mg of haloperidol per 0.5 ml or 5 mg of ziprasidone per 0.5 ml, whereas those who were 70 years of age or older received 0.25 ml of placebo or 1.25 mg of haloperidol per 0.25 ml or 2.5 mg of ziprasidone per 0.25 ml. Subsequent doses were administered every 12 hours at approximately 10 a.m. and 10 p.m. Research personnel doubled the volume and dose of the trial drug or placebo if a patient had delirium, was not yet receiving the maximum dose, and had not met criteria that required the trial drug or placebo to be withheld. Patients in the haloperidol group received a dose of up to 10 mg per administration and up to 20 mg per day, and those in the ziprasidone group received a dose of up to 20 mg per administration and up to 40 mg per day.
We halved the volume and dose of a trial drug or placebo if a patient did not have delirium (i.e., had a negative CAM-ICU assessment) for two consecutive assessments and was not yet receiving the minimum dose. We temporarily withheld a trial drug or placebo if a patient did not have delirium for four consecutive assessments or for safety reasons. We permanently discontinued a trial drug or placebo when any of the following occurred: torsades de pointes, neuroleptic malignant syndrome, drug reaction with eosinophilia and systemic symptoms syndrome, new-onset coma due to structural brain disease, or any life-threatening, serious adverse event that was related to the intervention, as determined by an independent data and safety monitoring board. We discontinued a trial drug or placebo after the 14-day intervention period or at ICU discharge, whichever occurred first.
To evaluate the efficacy and to guide volume and dose adjustments of trial drug or placebo, patients were assessed twice daily while they were receiving the intervention by trained research personnel using the CAM-ICU23,24 and the Richmond Agitation–Sedation Scale (RASS),25,26 a validated, 10-level scale that rates the level of consciousness from unresponsive to physical stimuli (score of −5) to combative (score of +4). We considered any day during which at least one CAM-ICU assessment was positive to be a day with delirium; a positive assessment was considered to indicate hyperactive delirium if the RASS score was higher than 0 and hypoactive delirium if the RASS score was 0 or lower.27
During the period when the patients were receiving a trial drug or placebo and for 4 days after discontinuation, we assessed the patients for side effects. Twice a day, before each administration of the intervention, we assessed the patients for a corrected QT prolongation of 550 msec using telemetry and, if telemetry indicated a corrected QT prolongation of 550 msec, we used 12-lead electrocardiography. Once daily, we assessed extrapyramidal symptoms using a modified Simpson–Angus Scale, a 5-item scale on which each item is scored from 0 to 4, with higher scores indicating worse extrapyramidal symptoms28; akathisia using a 10-point visual-analogue scale; and dystonia using a standardized definition.18
Treating clinicians were educated about the “ABCDE” treatment bundle (assess, prevent, and manage pain; both spontaneous awakening and breathing trials; choice of analgesia and sedation; assess, prevent, and manage delirium; and early mobility and exercise) and were encouraged to perform the treatment bundle to mitigate delirium among the patients in the ICU.29-33 Throughout the trial, we monitored its use and recorded adherence to each component of the bundle daily among the patients for whom informed consent was obtained.
The primary end point was days alive without delirium or coma (defined as the number of days that a patient was alive and free from both delirium and coma during the 14-day intervention period). Secondary efficacy end points included duration of delirium, time to freedom from mechanical ventilation (defined as extubation that was followed by at least a 48-hour period during which the patient was alive and free from mechanical ventilation), time to final successful ICU discharge (defined as the last ICU discharge during the index hospitalization that was followed by at least a 48-hour period during which the patient was alive and outside the ICU), time to ICU readmission, time to successful hospital discharge (defined as discharge that was followed by at least a 48-hour period during which the patient was alive and outside the hospital), and 30-day and 90-day survival. Safety end points included the incidence of torsades de pointes and neuroleptic malignant syndrome and the severity of extrapyramidal symptoms, as measured on the modified Simpson–Angus Scale.
Descriptions of sample-size calculations and statistical methods are included in the Supplementary Appendix. We sought to randomly assign 561 patients (187 per trial group), which would provide the trial with at least 80% power to detect a 2-day difference between groups in days alive without delirium or coma, at a two-sided significance level of 2.5% (after Bonferroni adjustment for two pairwise comparisons).
We analyzed all data using an intention-to-treat approach and compared the effects of haloperidol, ziprasidone, and placebo with respect to the primary end point using the Kruskal–Wallis test in unadjusted analyses and proportional-odds logistic regression in adjusted analyses. The primary analysis was performed with the use of an adjusted proportional-odds logistic-regression model that examined the effects of an intervention on days alive without delirium or coma, with a two-sided significance level of 2.5% to account for two pairwise comparisons, which were to be analyzed only if the P value for the overall effect across trial groups was significant. There was no plan for adjustment for multiple comparisons of secondary end points, and those results are reported without P values as point estimates with 95% confidence intervals that have not been adjusted for multiple comparisons. A total of 300 of 6100 (5%) potential assessments for delirium or coma were missing; we imputed these individual assessments using polytomous logistic regression that included multiple covariates. After calculating days alive without delirium or coma, we then used complete case analysis for all outcomes (see the Supplementary Appendix). We collected and managed data using REDCap electronic data-capture tools and used R software, version 3.4.4,34 for data management and statistical analyses. R code is available through Open Source Framework (https://osf.io/mq38r).