Association between delirium in the intensive care unit and subsequent neuropsychiatric disorders

We identified all adult ICU patients (≥ 18 years) who were admitted to one of the 14 medical-surgical ICU across Alberta, Canada (population 4.2 million people), between January 1, 2014, and June 30, 2016 [23]. Data was collected 5 years prior to an individual ICU admission and up to 1 year after hospital discharge to identify individuals who had a healthcare visit for a diagnosis of a neuropsychiatric disorder. Patients were excluded if their ICU stay was less than 24 h, if they died in ICU, if they did not have any delirium assessment in ICU, if they did not survive to hospital discharge, if their ICU admission did not link with hospital admission data within a 2-h window, or if their home was located outside of Alberta (to allow for complete follow-up). In the case of multiple ICU admissions during the study period, the patient’s first eligible ICU admission was used as the primary admission. Patients with an inpatient or outpatient healthcare visit for which any neuropsychiatric disorder (i.e., depressive, anxiety, trauma-and-stressor, and neurocognitive disorders) was listed as a post-admit comorbidity or a secondary diagnosis in the 5 years prior to and during their primary ICU admission were excluded (i.e., diagnostic codes suggestive of a pre-existing neuropsychiatric disorder). Hospital discharge dates ranged from January 9, 2014, to March 13, 2017. Patients were followed up until 1 year post-hospital discharge, death, or study end (October 31, 2017), whichever occurred first.

We used clinical and administrative databases from Alberta Health Services and Alberta Health (Discharge Abstracts Database [DAD], National Ambulatory Care Reporting System [NACRS], Physician Claims, Vital Statistics, and eCritical Tracer Database), previously successfully used in research studies [24, 25]. eCritical Tracer is a bedside electronic medical record that prospectively captures demographic, clinical, and patient outcomes for all patients admitted to an Alberta ICU [26]. The DAD captures patients who received hospital-based care and has up to 25 Canadian Enhancement of International Statistical Classification of Diseases, 10th Revision, fields for diagnostic codes. The NACRS captures data for all hospital-based and community-based emergency or ambulatory care and has up to 10 Canadian Enhancement of International Statistical Classification of Diseases, 10th Revision, fields for diagnostic codes. Physician Claims captures patients who received care from an outpatient clinic (primary or specialty) and has up to 3 Canadian Enhancement of International Statistical Classification of Diseases, 9th Revision, fields for diagnostic codes. Vital Statistics data is provided to Alberta Health Services from Alberta Health. All deaths occurring in Alberta must be registered with Alberta Vital Statistics. The DAD, NACRS, and Physician Claims were available until October 31, 2017, and Vital Statistics were available until December 31, 2017, at the time of data extraction.

The primary exposure was ever having delirium at any point during the ICU stay. Delirium was measured using the validated Intensive Care Delirium Screening Checklist (ICDSC) [27], which is documented as part of standard of care by the bedside nurse once every nursing shift (i.e., twice a day) [28]. The checklist is scored out of eight categories that include the level of consciousness, inattention, disorientation, hallucinations/delusions/psychosis, psychomotor agitation, inappropriate speech or mood, sleep wake/cycle disturbance, and symptom fluctuation [29]. Each category is coded as present (i.e., a score of 1) if the patient meets the criteria listed, for a maximum score of 8. Patients who scored ≥ 4 on the ICDSC at any time during the ICU stay were categorized as ever having delirium and those with all of their ICDSC scores < 4 on all ratings were categorized as never having delirium. The ICDSC has a sensitivity of 99% and a specificity of 64% in a medical-surgical ICU patient population [29]. Sensitivity analyses were conducted to determine whether there was a dose-response according to the duration of delirium in the ICU (one calendar day of delirium vs. two or more calendar days of delirium). One calendar day of delirium was determined by having at least one ICDSC score of ≥ 4 on that calendar day.

Neuropsychiatric disorders identified in previous systematic reviews to potentially be associated with delirium were included as outcome measures [30‐33], including depressive (such as major depressive disorders), anxiety (such as generalized anxiety disorders), trauma-and-stressor-related (such as acute stress disorders or post-traumatic stress disorder), and neurocognitive (such as dementia, respective symptoms for neurocognitive disorders include difficulty performing tasks in high stimuli environment, forgetting words, and experiencing personality changes [34]) disorders. Neuropsychiatric disorders were identified through physician-diagnosed ICD 9/10 codes that were captured in the DAD, NACRS, and Physician Claims databases (see Additional file 1). Validated coding algorithms were used to identify depressive [35] and anxiety [36] disorders. A neuropsychiatrist and psychiatrist developed coding algorithms for trauma-and-stressor-related and neurocognitive disorders as there are currently no published validated coding algorithms for these disorders. The two clinicians independently reviewed all DSM-5 codes as well as previous studies reporting these disorders identifying those ICD 9/10 codes for trauma-and-stressor-related and neurocognitive disorders. Disagreements were resolved through discussion. The primary outcome was a healthcare visit (e.g., hospital stay, emergency or ambulatory, and outpatient clinic) that was a post-admit comorbidity or a secondary diagnosis of a neuropsychiatric disorder (i.e., depressive, anxiety, trauma-and-stressor-related, and neurocognitive disorders). Neuropsychiatric disorders were examined together to identify individuals who were diagnosed with any of the four neuropsychiatric disorders and then for each disorder separately.

Characteristics potentially associated with delirium during the ICU stay and neuropsychiatric disorders at a 1-year follow-up were identified a priori [37, 38]. Characteristics included age, sex, ICU admission reason (medical, surgical, neurological, trauma), Acute Physiology and Chronic Health Evaluation (APACHE) II score, Charlson Comorbidity Index (CCI), Glasgow Coma Scale (GCS), transfer delay ≥ 24 h, invasive mechanical ventilation (yes/no), non-invasive mechanical ventilation (yes/no), continuous renal replacement therapy (yes/no), vasoactive medications (yes/no), ICU size (categorized as < 20 beds vs. ≥ 20 beds), teaching hospital (yes/no), ICU length of stay (categorized as < 7 days vs. ≥ 7 days), and last Sequential Organ Failure Assessment (SOFA) score.

Associations between delirium in the ICU and new neuropsychiatric disorders were assessed using log-binominal regression analyses (in which exponentiated regression coefficients represent risk ratios). Supplementary analyses were also conducted for each disorder separately (depressive, anxiety, trauma-and-stressor-related, and neurocognitive disorders) using log-binomial regression analyses. We adjusted all models for all covariates previously identified. Analyses were conducted in STATA version 14 (StataCorp LP; College Station, TX, USA) and R, version 3.5.1 (The R Project for Statistical Computing, http://​www.​r-project.​org/​).

From January 1, 2014, to June 30, 2016, there were 16,005 unique patients from the 14 study ICUs with at least one ICU stay. Of these ICU admissions, there were 11,166 unique ICU patients potentially eligible for inclusion in the study (exclusion reasons included ICU stay < 24 h, death in ICU, no delirium assessment, died in hospital following first ICU admission, patient’s ICU admission did not link with the DAD admission data within a 2-h window, non-Albertan patients). Of those, 6442 patients had a healthcare visit within the previous 5 years that included a diagnosis of a neuropsychiatric disorder, leaving 4033 patients who satisfied the inclusion criteria (Fig. 1). The baseline characteristics of patients with and without pre-existing neuropsychiatric disorders are outlined in Additional file 2. Among the 4033 patients included in the study, 475 died (12%) within the 1-year follow-up.

Patient and hospital characteristics of the primary study cohort are summarized in Table 1. The overall prevalence of delirium during the ICU stay was 44% (95% confidence interval [CI] 43–46%) (n = 1792). The median age of ICU patients in this study was 60 years (IQR 46–71 years), 68% (n = 2744) were male, the most common reason for an ICU admission was medical (e.g., pneumonia, cardiac arrest, sepsis, etc.), and the median length of ICU stay was 4.1 days (IQR 2.3–7.6 days). The median Charlson Comorbidity Index was 1 (IQR 0–3), median SOFA score upon ICU admission was 6 (IQR 3–9), and median APACHE II score upon ICU admission was 18 (IQR 13–24).

During the 1-year follow-up period, 794 patients (19.7%, 95% CI 18.5–20.9%) had a healthcare visit for a newly diagnosed neuropsychiatric disorder. The cumulative incidences of depressive, anxiety, trauma-and-stressor-related, and neurocognitive disorders were 10.1% (95% CI 9.2–11.1%), 8.8% (95% CI 8.0–9.7%), 2.9% (95% CI 2.3–3.4%), and 3.5% (95% CI 2.9–4.1%), respectively. The incidence density for those who developed any neuropsychiatric disorder during the 1-year follow-up was 213 (95% CI 200–226) per 1000 person years. The incidence densities of depressive, anxiety, trauma-and-stressor-related, and neurocognitive disorders were 110 (95% CI 100–120) per 1000 person years, 96 (95% CI 86–105) per 1000 person years, 31 (95% CI 25–36) per 1000 person years, and 38 (95% CI 32–44) per 1000 person years, respectively.

Patients who ever had delirium in the ICU had 1.14 (95% CI 0.98–1.33) times the risk of developing a neuropsychiatric disorder within 1 year compared to those who never had delirium in the ICU (Table 2). The adjusted risk ratios of depressive, anxiety, trauma-and-stressor-related, and neurocognitive disorders were 1.16 (95% CI 0.92–1.45), 1.16 (95% CI 0.92–1.47), 0.82 (95% CI 0.53–1.28), and 1.59 (95% CI 1.08–2.35), respectively, for those with delirium compared to those without delirium in the ICU (Table 2, see Additional file 3 for unadjusted estimates and Additional file 4 for adjusted estimates). A sensitivity analysis using duration of delirium, defined as (1) never having delirium, (2) one calendar day of delirium, and (3) two or more calendar days of delirium, showed similar adjusted risk ratios for any neuropsychiatric disorder, as well as each specific neuropsychiatric disorder of interest except neurocognitive disorders (see Additional file 5). Patients with two or more calendar days of delirium in the ICU had 2.14 (95% CI 1.40–3.28) times the risk of developing a neurocognitive disorder within the 1 year compared to those who never had delirium, whereas patients with one calendar day of delirium did not have a significantly different risk of developing a neurocognitive disorder within the 1 year compared to those who never had delirium (see Additional file 5).