Speech-based markers for posttraumatic stress disorder in US veterans

Area of Science:

• Psychiatry and Psychology
• Computational Linguistics
• Biomedical Engineering

Background:

• Current posttraumatic stress disorder (PTSD) diagnosis relies on subjective clinical interviews or self-report measures, leading to potential reporting inaccuracies.
• There is a need for objective diagnostic tools to accurately identify PTSD.
• Speech patterns may contain objective markers for PTSD diagnosis.

Purpose of the Study:

• To develop and validate an objective classifier for PTSD using speech features.
• To differentiate between warzone-exposed veterans with PTSD and healthy controls based on vocal biomarkers.

Main Methods:

• Speech samples were collected from 52 warzone-exposed veterans with PTSD and 77 controls, excluding those with major depressive disorder (MDD).
• A random forest (RF) algorithm analyzed 40,526 speech features from audio recordings of clinical interviews.
• The RF model was trained and tested to classify PTSD cases versus controls.

Main Results:

• The RF classifier achieved a high accuracy with an Area Under the Curve (AUC) of 0.954.
• The optimal classification achieved an 89.1% correct classification rate.
• Key speech features associated with PTSD included slower, more monotonous speech, and reduced tonal variation.

Conclusions:

• A speech-based algorithm can objectively distinguish individuals with PTSD from controls with high accuracy.
• The developed classifier demonstrates potential as an objective diagnostic tool for PTSD.
• Further validation and investigation into comorbid conditions like MDD are necessary.