Delirium and Benzodiazepines Associated With Prolonged ICU Stay in Critically Ill Infants and Young Children

Heidi A. B. Smith, MD, MSCI; Maalobeeka Gangopadhyay, MD; Christina M. Goben, MD; Natalie L. Jacobowski, MD; Mary Hamilton Chestnut, NP; Jennifer L. Thompson, MPH; Rameela Chandrasekhar, PhD; Stacey R. Williams, NP; Katherine Griffith, NP; E. Wesley Ely, MD, MPH; D. Catherine Fuchs, MD; Pratik P. Pandharipande, MD, MSCI

Disclosures

Crit Care Med. 2017;45(9):1427-1435. 

In This Article

Materials and Methods

We conducted a prospective observational study of pediatric patients, 6 months to 5 years old admitted to the PICU at Monroe Carell Jr. Children's Hospital at Vanderbilt regardless of diagnosis, for the following purposes: 1) to validate an objective delirium tool for critically ill infants and young children and 2) to determine associated risk factors and outcomes for delirium. The reliability and validity of the Preschool Confusion Assessment Method of the ICU (psCAM-ICU) for delirium monitoring in infants and young children have previously been published in Critical Care Medicine.[5] The data regarding delirium risk factors and outcomes presented in this article have not been previously published. Exclusion criterion for this study included the following: hearing/visual impairments, cognitive development less than expected for 6 months old, non-English speaking, planned PICU discharge/transfer, moribund, or for whom consent could not be obtained. The institutional review board approved this study. We assessed patients daily for delirium for up to 14 days or until transfer/discharge from the PICU, whichever came first. Delirium assessments were conducted only once daily due to restricted availability of research personnel for purposes of the validation study.

Baseline demographics and data pertaining to predisposing risk factors for delirium were collected upon admission including age, history of cyanotic cardiac disease, severity of illness (Pediatric Risk of Mortality III [PRISM III] score), and admission diagnosis. Additionally, information was collected regarding in-hospital risk factors which may contribute to the development and/or duration of delirium. These data included benzodiazepine and opioid exposures, cardiovascular Sequential Organ Failure Assessment (SOFA) score, need for mechanical ventilation, and presence of hypoxia. In patients with cyanotic heart disease, we defined hypoxia as oxygen saturations less than 75% or less than the patient's baseline saturations. Outcome data on both ICU and hospital LOS were also monitored and recorded. PRISM III is a severity of illness score that is calculated using the most abnormal variables during the 24 hours prior to and including admission to the PICU. PRISM scores range from 0 (best) to 30 (worst).[27] The cardiovascular SOFA score reflects degree of cardiovascular dysfunction in critically ill children, with a range from 0 (best) to 4 (worst), based on amount of vasopressor or inotropic support required. The cardiovascular SOFA is part of the modified SOFA score that has been shown to predict morbidity and mortality in critically ill children.[28]

Each day, patients were classified as either having "coma, delirium, or normal mental status." Level of arousal was first determined using a sedation scale, in this case, the Richmond Agitation Sedation Scale (RASS).[29] Patients who had a greatly decreased level of arousal to include either a RASS –4 or –5 were categorized as having "coma" and therefore were not assessed for delirium at that time, rather reassessed the following day. Patients who were spontaneously awake or aroused by voice (RASS >= –3) underwent delirium assessment (content of consciousness) using the psCAM-ICU. The psCAM-ICU is a highly valid and reliable bedside tool used by healthcare providers for delirium diagnosis in critically ill infants and young children (specificity, 91%; sensitivity, 75%; and κ, 0.79).[5] The psCAM-ICU assesses for key features of delirium: acute alteration or fluctuation from baseline mental status (feature 1), inattention (feature 2), acute alteration of consciousness (feature 3), and dysregulation of cognition/systems (feature 4). For "delirium" diagnosis using the psCAM-ICU, a patient must have a RASS score greater than or equal to –3 and demonstrate having features 1 and 2, plus either feature 3 or 4. A patient with a RASS greater than or equal to –3 without delirium was considered "normal."

Patient characteristics including demographics, admission diagnoses, the PICU clinical course, and outcomes were summarized using frequencies (%) for categorical variables and medians and interquartile ranges (IQRs) for continuous variables.

Multivariable Cox proportional hazard models were used to determine the associations between mental status (delirium, coma, or normal) and other a priori chosen risk factors (age, severity of illness, admission diagnosis of sepsis or acute respiratory distress syndrome [ARDS], cyanotic heart disease, benzodiazepine exposure, opiate exposure, dexmedetomidine exposure, cardiovascular SOFA score, and hypoxia) on the likelihood of ICU discharge. The hazard ratios (HRs) represent the chance of an event, or in these models, the likelihood of ICU discharge on a given day. A HR greater than one is positive or desirable as it correlates with a "greater likelihood" of ICU discharge on a given day and therefore a shorter LOS. Conversely, a HR less than one is negative or less desirable as it pertains to a "lower likelihood" of ICU discharge on a given day. All continuous variables were originally allowed to have a nonlinear relationship with outcomes using restricted cubic splines. If these nonlinear and/or interaction terms were clearly nonsignificant (p > 0.20) in a given model, they were removed from the final version in order to make interpretation and explanations as straightforward as possible. In other words, all covariates remained in the model as specified a priori; however, the level of complexity for the associations between covariates and the outcome was allowed to change. In all models, drug doses were converted using the cube root transformation in order to mitigate the influence of extremely high values.

We used negative binomial regression to determine the associations between delirium duration and both baseline and day 1 in-hospital risk factors (age, severity of illness, admission diagnosis of sepsis or ARDS, cyanotic heart disease, benzodiazepine and opiate exposures, cardiovascular SOFA score, mechanical ventilation, and hypoxia) that were present prior to the first delirium assessment. The outcome measure of delirium duration was the sum of daily positive delirium screenings for an individual patient, with up to 14 daily assessments or less if transfer/discharge from the PICU occurred prior to 14 days.

Finally, multinomial logistic regression was used to determine the role of risk factors (age, daily exposure to benzodiazepines and opioids, mechanical ventilation, cardiovascular SOFA score, and hypoxia) on mental status (normal, delirious, or comatose) from 1 day to the next. For purposes of this analysis, the outcome variable was the daily mental status assignments across the entire cohort. Since multiple daily assessments could be used from each hospitalization in this model, we used bootstrapping clustered by patient to adjust the SEs and account for correlation among assessments from the same patient.

All analyses were performed using R-statistical software version 3.3.0 (R Foundation for Statistical Computing, Vienna, Austria; http://www.r-project.org). We graphically assessed the proportional hazard assumption for time-to-event model(s); these assumptions were met satisfactorily.

Comments

3090D553-9492-4563-8681-AD288FA52ACE

processing....