Development and Validation of Simplified Delirium Prediction Model in Intensive Care Unit

Abstract

Background: The intensive care unit (ICU) is where various medical staffs and patients with diverse diseases convene. Regardless of complexity, a delirium prediction model that can be applied conveniently would help manage delirium in the ICU. Objective: This study aimed to develop and validate a generally applicable delirium prediction model in the ICU based on simple information. Methods: A retrospective study was conducted at a single hospital. The outcome variable was defined as the occurrence of delirium within 30 days of ICU admission, and the predictors consisted of a 12 simple variables. Two models were developed through logistic regression (LR) and random forest (RF). A model with higher discriminative power based on the area under the receiver operating characteristics curve (AUROC) was selected as the final model in the validation process. Results: The model was developed using 2,588 observations (training dataset) and validated temporally with 1,109 observations (test dataset) of ICU patients. The top three influential predictors of the LR and RF models were the restraint, hospitalization through emergency room, and drainage tube. The AUROC of the LR model was 0.820 (CI 0.801–0.840) and 0.779 (CI 0.748–0.811) in the training and test datasets, respectively, and that of the RF model was 0.762 (CI 0.732–0.792) and 0.698 (0.659–0.738), respectively. The LR model showed better discriminative power (z = 4.826; P < 0.001). Conclusion: The LR model developed with brief variables showed good performance. This simplified prediction model will help screening become more accessible.