Derivation of a Clinical Prediction Rule for Sedative-Associated Delirium During Acute Respiratory Failure
Year: 2021
Type: conference-abstract
Abstract: Rationale: Recent work has shown that delirium, which affects up to 80% of mechanically ventilated ICU patients and is associated with adverse outcomes, can be categorized by clinical phenotype. Sedative-associated delirium is the most common and likely the most modifiable phenotype since clinicians determine which patients are sedated and which sedative(s) a patient receives. A tool that predicts risk for sedative-associated delirium could aid clinicians when selecting a sedative agent. We, therefore, sought to generate a predictive model that could form the basis for future work examining the effectiveness of personalized, clinical prediction rule-directed sedation in the ICU. Methods: Using data from the multicenter, observational, prospective BRAIN-ICU cohort study, which recruited adult ICU patients with acute respiratory failure and/or shock, we analyzed all participants who received a benzodiazepine, opioid, propofol, and/or dexmedetomidine. Participants were assessed for delirium at least once daily using the CAM-ICU, and we considered sedativeassociated delirium to be present when the CAM-ICU was positive on a day when the participant received a sedative. We derived a predictive model of sedative-associated delirium within 72 hours of study enrollment using multivariable backwards logistic regression with an exclusion threshold of p = 0.20. We selected 29 candidate variables based on previous work identifying risk factors for delirium, biological plausibility, and widespread availability in non-research settings. Results: The 977 patients analyzed had a median age (interquartile range) of 63 (53-72) years old and a median Sequential Organ Failure Assessment (SOFA) score of 9 (7-12) at admission. The most common ICU admission diagnoses were sepsis/ARDS (31%), cardiac disease (17%), and surgery (16%). Age, history of cancer, some ICU admission diagnoses, severity of illness (as measured by SOFA score at study enrollment), and mechanical ventilation were significant predictors included in the model. Receipt of a benzodiazepine, opioid, and propofol were each associated with increased odds of sedative-associated delirium, whereas receipt of dexmedetomidine was not associated and therefore not included in the model. The generated model had good predictive value, with an area under the receiver-operator curve of 0.83. Conclusions: A predictive model for sedative-associated delirium derived from BRAIN-ICU cohort data is highly predictive, with an AUROC similar to those of other widely-used clinical prediction rules (e.g. the Wells score). If externally validated, the model may be useful as a clinical prediction rule that guides sedative practice, particularly for patients at high risk of sedative-associated delirium.
Institutions: University of Pittsburgh, Vanderbilt University
Language: English
Field: Medicine
Domain: Health Sciences
Open Access status: gold
OpenAlex ID: W4234463081