Predicting Traumatic Brain Injury Post-Trauma Using Temporal Attention on Sleep-Wake Data

Area of Science:

• Neurology
• Biomarkers
• Artificial Intelligence

Background:

• Traumatic Brain Injury (TBI) presents a significant public health challenge.
• Accurate TBI classification is crucial for effective patient treatment and outcomes.
• Sleep/wake behavior is a potential biomarker for TBI, but optimal detection windows are unknown.

Purpose of the Study:

• To determine the optimal time window for utilizing sleep/wake data to classify Traumatic Brain Injury (TBI).
• To evaluate the impact of data collection duration and time from trauma on TBI classification accuracy using deep learning.

Main Methods:

• A deep learning model was employed to analyze longitudinal sleep/wake data from over 2,000 emergency department patients.
• The study included patients with and without blood biomarker-documented TBI (Glial Fibrillary Acidic Protein - GFAP > 268 pg/ml).
• The model assessed the influence of time from trauma and data collection duration on distinguishing TBI+ from TBI- cases.

Main Results:

• Sleep/wake data from the first 7 days post-TBI demonstrated the highest accuracy in identifying TBI.
• Sensitivity and specificity for the first 7, 14, and 21 days were 81%/25%, 40%/66%, and 45%/58%, respectively.
• Deep learning models using data from the first 7, 14, and 21 days achieved F1 scores of 22%, 21%, and 20%, respectively.

Conclusions:

• Early sleep/wake data shows significant promise for aiding in Traumatic Brain Injury (TBI) identification.
• Future TBI diagnostic tools could incorporate sleep/wake biomarkers for enhanced screening and intervention.