Wearable Detection of Divergent Anxiety Types: Comparing Physiological Signatures and Shared Arousal Patterns Using Machine Learning

Background: Contamination Obsessive-Compulsive Disorder (OCD) anxiety and needle anxiety are distinct but functionally impairing anxiety subtypes. They differ in their autonomic signatures, with contamination OCD associated with sympathetic hyperactivation and needle anxiety associated with vasovagal responses. Despite these differences, both respond to exposure-based therapy and share experiential features that may give rise to common physiological patterns.Objectives: This study examined how physiological markers collected from a wearable device differ across contamination OCD and needle anxieties, and whether shared psychophysiological features can nonetheless be identified. It also assessed whether anxiety states and severity levels can be accurately detected in both groups. Methods: 48 participants (16 with contamination OCD anxiety and 32 with needle anxiety) were exposed to virtual reality scenarios designed to elicit anxiety. heart rate (HR), electrodermal activity (EDA), and skin temperature were collected using an Empatica E4 wearable. K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Random Forest, and Decision Tree machine learning (ML) models were trained to classify anxious versus baseline states (Experiment 1) and symptom severity levels (Experiment 2), using 10-fold cross-validation. Results: Skin temperature was the most accurate single marker across both anxiety types, with classification accuracy improving when combined with EDA. KNN consistently out-performed other ML models, reaching up to 99.3% accuracy and 1.0 F1 score. While autonomic patterns diverged across symptom types, overlapping physiological features, particularly skin temperature and EDA, were observed in both. Conclusions: Physiological data from wearables can effectively detect anxiety and differentiate symptom severity across symptomatically and physiologically distinct presentations. These results suggest that specific physiological markers, especially skin temperature and EDA, may reflect shared arousal processes relevant to anxiety across diagnostic boundaries. The findings support the development of scalable, passive tools for early detection and monitoring of anxiety using wearable technologies.