Acute Stress Disorder (ASD) is a psychological response to a terrifying, traumatic, or surprising event and can be a precursor to Post Traumatic Stress Disorder (PTSD) if left untreated. The current method of diagnosing ASD is anecdotal in practice, with clinical evaluation involving patient history and physical examination within days following a traumatic event. Diagnostic tools typically include screenings and questionnaires, but they require significant training and often miss initial physiological symptoms. The addition of Artificial Intelligence (AI) data-driven decision support using electroencephalography (EEG) and heart rate monitoring of biopotential activity may aid in detecting early signs of trauma-related disorders.
To test this hypothesis, participants from Greenville Memorial Hospital Emergency Department and Level 1 Trauma Center wore an EEG headset prior to their work shift to determine a reference signal (preliminary sample size of ~n=20). Participants then wore the EEG headset during their work shift while the timing of traumatic events and severity was recorded and biopotential data was collected and stored. De-identified data was exported and processed into various waveforms before being fed into an AI model to develop parameters to identify increased stress levels.
The Acute Stress Center (ASC) developed an application that enables physicians to drag and drop EEG data into the application, select which waveforms are of interest to their research, and view summary statistics of their data and graphs indicating stressful events along their EEG waveform. In addition, ECG data is displayed alongside the EEG data to allow a visual comparison.
The prototype and application were both further enhanced to gather and evaluate heart rate in addition to EEG data. Data was collected outside of a hospital setting with lower stress levels to test the algorithm. From preliminary testing, the addition of heart rate did not show much significance in predictors of traumatic responses, ranking second to last in feature importance. Furthermore, among both trials, the GammaMid waveform still proved to be the most important waveform in the indication of a traumatic response with a score of 100%. Additionally, correlations between waveforms and heart rate were fairly insignificant with the highest correlation of .26. Similarly to testing with only EEG waveforms, a plateau in accuracy occurs when 6 variables are included. However, in EEG only trials, accuracies of 95% were achieved, while in the heart rate trial a maximum accuracy of 80% could be achieved. Discrepancies in data could be attributed to the fact that data was collected in stressful situations rather than traumatic due to preliminary testing of the prototype before gathering data at Prisma Health. Variance in results may be seen in more high stress situations and will be evaluated in future work.