The React & Rebound Model: Capturing Emotion Regulation Dynamics from Passive Wearable Data

Background Emotion regulation—the ability to respond to and restore equilibrium after emotional perturbations—is central to mental health. Yet objective measurement remains limited to lab-based studies with group-level results, while consumer wearables focus on physical activity-related metrics rather than emotional dynamics.

We aimed to develop computational models that extract personalized, interpretable emotion regulation parameters from continuous heart rate variability (HRV) data collected via consumer wearables during everyday life, and validate these parameters against self-reported anxiety symptoms.

We analyzed 4 weeks of continuous HRV data from N = 49 healthy adults wearing Samsung Galaxy Active 2 smartwatches. We derived a continuous autonomic balance signal and developed three computational modeling approaches of increasing sophistication: (1) a static sympathetic load metric, (2) an Ornstein-Uhlenbeck (OU) dynamical systems model capturing continuous restoration dynamics, and (3) a discrete-state Markov transition model—the React & Rebound model— capturing reactivity and rebound dynamics. All models were estimated using joint hierarchical Bayesian models that simultaneously extract subject-specific parameters from HRV time series and estimate their association with Generalized Anxiety Disorder 7-item scale (GAD-7) scores. The validity of extracted parameters was evaluated against anxiety symptom severity.

Static sympathetic load correlated modestly with GAD-7 (r = 0.39, R2 = 0.16). The OU model captured 69% of variance (R2 = 0.69), and the React & Rebound model captured 60% (R2 = 0.60) with substantially fewer parameters. Both models revealed that anxiety symptom severity is associated with the interaction between activation and restoration parameters—not either alone. Fast rebound appeared protective even for highly reactive individuals, who scored comparably to low-reactivity groups when restoration was rapid (Cohen’s d = 1.17 between highest- and lowest-risk quadrants). In the OU model, the interaction effect was specific to GAD-7 scores versus PHQ-9 and ISI scores; in the React & Rebound model, the interaction was credible across all three symptom measures. Both models were unchanged after controlling for physical activity (ΔR2 < 0.002).

Computational models can extract interpretable emotion regulation parameters from naturalistic wearable data. The React & Rebound model yields two personalized parameters—reactivity and rebound—that are strongly associated with anxiety symptoms and define meaningful autonomic profiles. These parameters bridge autonomic dynamics measurable via consumer devices to neural circuit models of emotion regulation, with implications for characterizing individual autonomic profiles via consumer wearables. Competing Interest Statement The authors have declared no competing interest. Abbreviations - bpm - beats per minute - ESS - effective sample size - GAD-7 - Generalized Anxiety Disorder 7-item scale - HDI - highest density interval - HR - heart rate - HRV - heart rate variability - IBI - inter-beat interval - ISI - Insomnia Severity Index - MCMC - Markov chain Monte Carlo - NUTS - No-U-Turn Sampler - OU - Ornstein-Uhlenbeck - PHQ-9 - Patient Health Questionnaire 9-item scale - PPG - photoplethysmography - PSIS-LOO - Pareto-smoothed importance sampling leave-one-out - RMSSD - root mean square of successive differences - SEM - standard error of measurement - vmPFC - ventromedial prefrontal cortex