Presentation Information
[SY-57-03]Enhancing Early Treatment Response Prediction in Panic Disorder Using a Virtual Reality-Based Assessment Tool: Integrating Multimodal Indicators with Machine Learning
*Junhuyng Kim (Department of psychiatry, Samsung kangbuk Hospital, Sungkyunkwan University School of Medicine, Seoul, Republic of Korea(Korea))
Keywords:
virtual reality,panic disorder,early treatment response,machine learning
Background: Early treatment response (ETR) is a robust predictor of long-term outcomes in anxiety disorders, including panic disorder (PD). However, conventional assessments may lack ecological validity and sensitivity to early psychophysiological changes, limiting their utility in real-world clinical settings. Objective: We aimed to evaluate the predictive potential of the Virtual Reality Assessment of Panic Disorder (VRA-PD)—a novel VR-based tool capturing subjective and physiological responses during anxiety-provoking scenarios—for identifying ETR in patients with PD. Methods: Fifty-two individuals (25 PD patients and 27 healthy controls [HCs]) completed assessments every two months for six months. Measures included VR-based anxiety scores, heart rate variability (HRV), conventional clinical scales (e.g., Panic Disorder Severity Scale, Anxiety Sensitivity Index), and demographic variables. PD patients were categorized as early responders (ER, n = 7) or delayed responders (DR, n = 18) based on symptom change trajectories. Results: A CatBoost machine learning model incorporating both VR-based and conventional features showed improved performance in classifying ER, DR, and HCs (accuracy: 85%, F1-score: 0.71), outperforming models using only conventional (accuracy: 77%, F1-score: 0.56) or VR-only (accuracy: 75%, F1-score: 0.64) data. Performance further improved when restricted to the top 10 predictors identified by SHapley Additive exPlanations (accuracy: 90%, F1-score: 0.83). Key features included VR-based anxiety responses, HRV indices, and clinical severity ratings. Conclusions: The integration of immersive VR-based assessment and machine learning enables accurate ETR prediction in PD, addressing key limitations of conventional methods. These findings support the clinical utility of digital phenotyping and VR technologies in developing personalized, ecologically valid treatment strategies in anxiety disorders.