Bio-Signal Based Detection of Hyperarousal as Onset of Post-Traumatic Stress Disorder (PTSD) Episodes

Post-Traumatic Stress Disorder (PTSD) affects roughly 314 million individuals globally and 13 million Americans each year. The condition is characterized by sporadic manifestation of hyperarousal periods known as PTSD episodes. Untreated PTSD can lead to the worsening of such episodes and is a leading factor for the high suicide rate among PTSD patients. While screenings and diagnoses for the condition are available at clinics, there is no real-time detection system for these episodes, so they tend to go undetected and untracked. This study aimed to develop and test a proof-of-concept tool using machine learning algorithms to predict and detect Hyperarousal episodes from bio-signal data collected from wearables. The bio-signals used were heart rate variability (HRV), and respiration rate (RR), as both are collected and tracked on common wearables. A machine learning classification model with logistic regression was developed using two separate datasets from the MIT PhysioNet database, each comprised of biomarker data collected through wearable devices during both resting and hyperarousal states. The initial biomarker observations before stimulation/hyperarousal were used as a baseline and were later used to calculate the fluctuations. The logistic regression model gave sound classification results and correctly identified a hyperarousal state (imitating PTSD episodes) 92.1% of the time. This model was validated and had a precision of 81.7% in classifying episodes in both datasets and had an F1 score of 86.59. These results indicate that classification models based on HRV and RR can be used for detecting a patient’s hyperarousal state associated with PTSD.