PREDICTION OF DELIRIUM, MORTALITY, AND FALL RISK IN INPATIENTS USING BISPECTRAL EEG

Kasra Zarei; Gen Shinozaki

doi:10.1016/j.jagp.2019.01.126

Delirium is a prevalent yet under-diagnosed and under-treated condition. Although questionnaire instruments are capable of detecting delirium if implemented rigorously, they have not been effectively used due to their lack of efficiency in busy hospital workflows. Their subjective nature is also a drawback. Electroencephalography (EEG) can objectively detect the “diffuse slowing” of brain waves that is characteristic of delirium. While standard EEG is not suitable for mass screening due to its size, cost, and the expertise required for lead placement and interpretation, simplified two channel EEG devices can still be used to measure frontal EEG changes. Our goal was to investigate if frontal EEG activity could be used to detect delirium and to predict fall risk and mortality among elderly inpatients. A single center, prospective design was used to collect frontal EEG activity (Fp1 and Fp2 EEG signals) from elderly patients (>55 yo) after admission or at the time of an emergency room visit. EEG features (band powers and different combinations of frequency bands) were calculated for both channels and averaged. A bispectral EEG score (“BSEEG score”) was calculated based on the distribution of the scores from the study participants, and normalized with a mean value of zero and one standard deviation (SD) above the mean as 1. The top 9 EEG features were selected using machine learning methods, Random Forest. Subjects were assessed for the clinical presence of delirium and the primary outcomes measured were fall history, and mortality. K-nearest neighbors, logistic regression, support vector machine (SVM), kernelized SVM, and neural network approaches were used to assess the ability of bispectral EEG to predict delirium status, falls and survival. EEG features and outcome data for 274 patients was available for analysis. Clinical presence of delirium and BSEEG score were significantly associated (P = 6.14 × 10-6; unadjusted, P= 1.17 × 10-5; adjusted). Of all the classification methods, kernelized SVM yielded the highest prediction accuracies of 69%, 89%, and 81%, for delirium status, falls and mortality, respectively. Hazard ratio (HR) for survival controlling for age, gender, CCI and delirium status based on one SD change of BSEEG score at the time of admission was 1.32, (CI = 1.03 to 1.70, P=0.026). In our study, bispectral EEG in delirious elderly inpatients was able to predict patient outcomes including mortality and fall risks. Bispectral EEG monitoring may also be applicable in settings including the primary care clinic, emergency department, and in nursing home or home-care settings. Delirium is particularly dangerous when patients experience it outside of hospitals because they do not have medical attention available on site. The simple, noninvasive nature of a simplified EEG test makes it potentially ideal for routine screening. Potentially it can be used as a next vital sign to monitor the risk of delirium, mortality, and falls among elderly patients. As the aging population is expanding rapidly, such a test would be in high demand. This study was supported by the University of Iowa Research Foundation GAP funding award for Gen Shinozaki. Gen Shinozaki has grant support from NSF1664364 and K23 MH107654.

PREDICTION OF DELIRIUM, MORTALITY, AND FALL RISK IN INPATIENTS USING BISPECTRAL EEG

View Online

Abstract

Details

Metrics