Dynamics of Multimodal Stress Biomarkers: ECG, EDA, and PPG across various conditions

Lukas Smital1, Martin Vitek2, Marina Filipenska2, Radovan Smisek1, Lucie Saclova2, Andrea Nemcova1
1Brno University of Technology, Faculty of Electrical Engineering and Communication, Department of Biomedical Engineering, 2Brno University of Technology


Abstract

Aims: Real-time stress detection using wearable sensors requires understanding which physiological modality responds first. Most studies compare stress and baseline as static epochs, thereby discarding temporal information. We investigated the temporal evolution of ECG, EDA, and PPG stress biomarkers across multiple conditions, including anticipatory phases preceding the stressor. Methods: We analyzed the WESAD dataset (15 subjects) recorded during the Trier Social Stress Test, including baseline, stress, amusement, and meditation conditions. Signals included chest ECG (700 Hz), wrist EDA (4 Hz), and wrist PPG (64 Hz). Six features were selected: heart rate (HR) and short-term HR variability (RMSSD) from ECG; tonic skin conductance level (meanSCL) and skin conductance response rate (SCR) from EDA and normalized pulse area and pulse skewness derived from PPG. Using a sliding-window approach (30 s window, 5 s step), we computed paired Cohen's d relative to a 5-minute resting baseline for each condition, including preceding transitions period. Results: ECG features showed the earliest response, with HR rising already during the anticipatory transition preceding stress onset (d = 1.6). RMSSD decreased (d = -0.9), reflecting vagal withdrawal. EDA features responded more slowly, with meanSCL reaching d = 0.9 and SCR d = 1.6 primarily during the stress phase itself. PPG features showed strong effects (pulse area d = -2.3, skewness d = 1), but with higher variability, likely due to motion sensitivity of wrist PPG. Meditation produced a mirror-image response in HR, RMSSD, SCR and pulse area compared to stress, consistent with vagal recovery. Amusement and baseline showed minimal effects (|d| < 0.3), confirming stress specificity. Conclusion: ECG-derived features capture stress responses earlier than EDA, emerging already during the anticipatory phase. ECG-based features appear most suitable for early wearable stress detection, while EDA provides slower but complementary information. Minimal effects during amusement and baseline confirm stress specificity.