Decoding accelerometry for classification and prediction of critically ill patients with severe brain injury

Area of Science:

• Neurology
• Biomedical Engineering
• Intensive Care Medicine

Background:

• Severe brain injury (SBI) presents challenges in monitoring neurological status and predicting patient outcomes.
• Quantitative motor features are underexplored in critically ill patients with SBI.

Purpose of the Study:

• To explore quantitative motor features in critically ill patients with SBI.
• To determine if computational decoding of motor features can inform neurological states and outcomes.
• To assess the utility of wearable microsensors for continuous motor data acquisition.

Main Methods:

• Prospective cohort study of 69 ICU patients with SBI.
• High-frequency accelerometry data collected from all extremities using wearable microsensors (median 24.1 hours/patient).
• Machine learning models trained on time-, frequency-, and wavelet-domain features to detect responsiveness (Glasgow Coma Scale motor score, GCSm) and predict functional outcome (Glasgow Outcome Scale-Extended, GOSE).

Main Results:

• Detection models accurately discriminated patients capable of purposeful movement (GCSm > 4) with significant AUC (0.70) across observation windows from 12 minutes to 9 hours.
• Prediction models accurately discriminated patients with upper moderate disability or better (GOSE > 5) within 2-6 hours of observation (AUC: 0.82).
• Time series analysis of motor activity provided clinically relevant insights into functional states and short-term outcomes.

Conclusions:

• Quantitative motor features derived from wearable accelerometry are valuable for assessing neurological status in SBI patients.
• This approach offers a non-invasive method for predicting functional outcomes in the short term.
• Wearable sensor technology holds promise for improving patient management and prognostication in intensive care settings for SBI.