Study design
We performed a secondary analysis on data collected of 5057 ICU patients for a prospective multicenter implementation study [ICU Delirium in Clinical Practice Implementation Evaluation (iDECePTIvE) study] [10]. The iDECePTIvE study was aimed to optimize adherence to delirium-oriented measures and was performed between 2012 and 2015 in six ICUs in the Rotterdam area, The Netherlands [10]. The implementation was constructed in three phases: after a baseline assessment (phase I), delirium assessment was implemented in all ICUs using either the Confusion Assessment Method for ICU (CAM-ICU) [11] or Intensive Care Delirium Screening Checklist (ICDSC), based on local preference [12]. Each participating ICU used one of these delirium assessment methods consistently and non-interchangeably. This baseline phase was followed by three further measurement periods (phases II–IV), focusing on the effects of implementation on adherence to daily assessments for delirium and to presence of other guideline recommendations from the Pain, Agitation and Delirium (PAD) guidelines issued by the Society of Critical Care Medicine [13]. Thus, phases II–IV implicated well embedded and structured three-times daily delirium assessments by ICU nurses.
After implementation of the PAD guideline recommendations [13], the iDECePTIvE study found improvements in delirium screening, the degree of physiotherapy and early mobilization, and the use of light sedation in ventilated patients [10]. The authors also reported a decrease in the use of benzodiazepines and delirium and coma duration.
The prospective data registry from the iDECePTIvE study was approved by the Medical Ethical Committee of the Erasmus University Medical Center (registration number: MEC-2012-063) and was not subjected to the Dutch law ‘Medical Research Involving Human Subjects’ (WMO). As such, the need for informed consent was waived. Involved investigators handled and analyzed anonymized data according to Dutch regulations. This study was reported using the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement [14].
Study population
Inclusion criteria for this secondary analysis were adults (≥ 18 years) admitted to the medical or surgical ICU in phases II–IV, who had daily delirium assessments with either the CAM-ICU or ICDSC, and whose level of sedation was assessed daily with the Richmond Agitation–Sedation Scale (RASS) [15]. Patients with a primary neurological admission diagnosis had been excluded. Additional exclusion criteria were persisting coma (defined as RASS score − 4 or − 5 during the entire ICU stay) and ICU readmissions.
Outcomes
The primary outcome was hospital mortality. Secondary outcomes were ICU mortality, ICU length of stay (in days), presence of coma during ICU stay (yes/no) and number of coma days, use of mechanical ventilation during ICU stay (yes/no) and number of ventilation days. Furthermore, we related the presence of delirium and the delirium subtypes to various management variables to gain further insight into their phenotypes: use of antipsychotics (yes/no) and number of days with haloperidol administration and haloperidol dose, and continuous intravenous administration during ICU stay of sedatives (yes/no), benzodiazepines (yes/no) and opioids (yes/no), and number of days of administration.
Data collection
Patients were followed from ICU admission until hospital discharge. The following demographic data were prospectively collected in the Case Report Form or in the electronic patient data management system: age, sex, Acute Physiology and Chronic Health Evaluation (APACHE) IV score, ICU admission diagnosis (medical, elective surgery, or acute surgery), ICU length of stay, and ICU mortality and hospital mortality. Further, during ICU stay the following daily data were collected: mechanical ventilation, delirium assessments with CAM-ICU or ICDSC (three times daily), RASS scores, administration of antipsychotics [yes/no, including haloperidol (with dose), olanzapine, or quetiapine], and continuous intravenous administration during at least 2 h/day of sedatives (yes/no, including clonidine, dexmedetomidine or propofol), benzodiazepines (yes/no, including midazolam or lorazepam) and opioids (yes/no, including morphine, fentanyl or remifentanil).
Definitions
We defined patients with ICU delirium as those with at least one positive delirium assessment with CAM-ICU or ICDSC during their ICU stay. We classified the delirium subtypes based on the combined results of the delirium assessment with either the CAM-ICU or ICDSC and the RASS score [15]. The RASS score ranges from − 5 (unarousable) to + 4 (combative), in which a RASS score of 0 indicates that the patient is calm and alert. If the RASS score was − 4 or − 5, delirium assessment was not possible [11]. Delirium was assessed with the CAM-ICU in three participating ICUs and in the fourth ICU the ICDSC was used. Delirium assessment was performed three times daily (once per 8-h shift) by ICU nurses, previously trained to use the CAM-ICU or ICDSC (depending on the hospital) [10, 16]. Hyperactive delirium was defined as a persistently positive RASS score (+ 1 to + 4) during all positive delirium assessments throughout the entire ICU stay, whereas a persistently negative or neutral RASS score (0 to − 3) at each positive delirium assessment was defined as hypoactive delirium [17]. Mixed delirium was defined as both hyper- and hypo-active delirium during ICU stay.
Further, regarding the secondary outcomes, the use of mechanical ventilation (yes/no) was defined as at least one mechanical ventilation day during ICU stay. Similarly, coma during ICU stay (yes/no) was defined as at least one coma day, with a coma day being a day on which patients had a RASS score of − 4 or − 5, but never obtained a RASS score of − 3 or higher, hindering delirium assessment. If patients had received antipsychotics (haloperidol, quetiapine, olanzapine), or continuous intravenous sedatives (clonidine, dexmedetomidine, propofol), benzodiazepines (midazolam or lorazepam) or opiates (morphine, fentanyl, remifentanil) for ≥ 2 h/day during ICU stay, they were defined accordingly (yes/no). Mean haloperidol daily doses were only reported for days on which patients received haloperidol.
Statistical analyses
Continuous data were summarized as means with standard deviations or as medians with interquartile ranges (IQR), depending on distribution. Categorical variables were shown in frequencies and percentages. To assess differences between non-delirious patients and delirious patients, and between the delirium subtypes, χ2 tests were used for categorical data, independent t-tests for continuous normally distributed variables, and Mann–Whitney U tests for continuous non-normally distributed variables.
For the primary analysis, we used mixed-effects logistic regression models, with adjustment for prognostic factors related to delirium and mortality (APACHE IV, age, and ICU admission diagnosis) [18,19,20], and a random intercept for hospital. We tested for the interaction between delirium subtype and the APACHE IV score with the likelihood ratio test (LRT) and added the interaction term to the model when significant. Further, a propensity score model was used to match patients of the different subtypes with non-delirious patients. The propensity of having a specific delirium subtype was estimated using a mixed-effects multivariable logistic regression analysis with delirium subtype (yes/no) as an outcome, and the same independent variables as the primary logistic regression analysis. Patients with non-overlapping propensity scores (calliper of max 10%) were excluded for the effect analyses. Additionally, we performed a sensitivity analysis in order to investigate the interaction of delirium subtypes and concurrent guideline-based delirium-oriented measures, e.g., physiotherapy and early mobilization, and vice versa, by excluding patients who were included in phase II (i.e., the phase in which only delirium screening was implemented and not the guideline implementation measures including those related to physiotherapy and early mobilization).
For the secondary outcomes, logistic mixed-effects regression analyses were used to study ICU mortality, presence of coma, use of mechanical ventilation, antipsychotics, continuous intravenous sedatives, benzodiazepines and opioids, and linear mixed-effects regression analyses for ICU length of stay, number of delirium days, coma days and ventilation days, and the number of days on which antipsychotics (including haloperidol dose), sedatives, benzodiazepines or opioids were applied. We adjusted for the APACHE IV score, age, and ICU admission diagnosis, with a random effect for hospital. An interaction term of delirium subtype and APACHE IV score was added to the model when significant with the LRT.
A p-value < 0.05 was considered statistically significant. Descriptive analyses were performed using SPSS, version 24. Logistic and linear regression analyses and propensity score matching were performed in the R statistical software (version 4.1.0), for which multiple imputation was used to handle missing values, with use of the mice package in R [21].