Delirium detection using wearable sensors and machine learning in patients with intracerebral hemorrhage

• DOI: 10.3389/fneur.2023.1135472
• Publication Date: 2023-06-09T05:05:54Z
• Authors: Abdullah Ahmed; Augusto Garcia-Ag...; Ivana Petrovic; Fatemeh Radaei; James Fife; John Zhou; Hunter Karas; Scott Moody; Jonathan Drake; Richard N. Jones; Carsten Eickhoff; Michael E. Reznik
• Keywords: Neurointensive care; Stroke; Hemiparesis

Delirium is associated with worse outcomes in patients stroke and neurocritical illness, but delirium detection these can be challenging existing screening tools. To address this gap, we aimed to develop evaluate machine learning models that detect episodes of post-stroke based on data from wearable activity monitors conjunction stroke-related clinical features. Prospective observational cohort study. Neurocritical Care Stroke Units at an academic medical center. We recruited 39 moderate-to-severe acute intracerebral hemorrhage (ICH) hemiparesis over a 1-year period [mean (SD) age 71.3 (12.20), 54% male, median (IQR) initial NIH Scale 14.5 (6), ICH score 2 (1)]. Each patient received daily assessments for by attending neurologist, while were recorded throughout each patient's hospitalization using wrist-worn actigraph devices (on both paretic non-paretic arms). compared the predictive accuracy Random Forest, SVM XGBoost methods classifying status information alone combined data. Among our study cohort, 85% (n = 33) had least one episode, 71% monitoring days 209) rated as delirium. Clinical low detecting day-to-day basis [accuracy mean 62% (18%), F1 50% (17%)]. Prediction performance improved significantly (p < 0.001) addition 74% (10%), 65% (10%)]. actigraphy features, night-time especially relevant classification accuracy. found improves stroke, thus paving way make actigraph-assisted predictions clinically actionable.