Minimizing Survey Questions for PTSD Prediction Following Acute Trauma

Area of Science:

• Psychiatry
• Computational Neuroscience
• Machine Learning

Background:

• Traumatic events can lead to post-traumatic stress disorder (PTSD), impacting social and occupational functioning.
• Machine learning models show promise for predicting PTSD development from assessments.
• Current methods often involve lengthy questionnaires, posing administrative challenges.

Purpose of the Study:

• To develop a predictive model for PTSD 3 months post-trauma.
• To minimize the number of survey questions required for accurate PTSD prediction.
• To maintain high prediction accuracy using a reduced set of questions.

Main Methods:

• Formulated PTSD prediction as a feature selection problem.
• Evaluated four distinct feature selection approaches.
• Utilized survey-based psychological assessments administered within 2 weeks post-trauma.

Main Results:

• Achieved up to 72% accuracy in predicting 3-month PTSD diagnosis.
• Successfully predicted PTSD using only 10 survey questions.
• Employed a mean decrease in impurity feature selector and a gradient boosting classifier.

Conclusions:

• It is feasible to accurately predict PTSD development with a significantly reduced number of survey questions.
• Machine learning, specifically feature selection, can streamline PTSD risk assessment.
• This approach offers a time-efficient alternative for early PTSD identification in trauma patients.