A Wearable-Derived Coherence State Model for Early Warning and Recovery Profiling of Stress-Related Dysregulation V1.0

This record contains the manuscript, A Wearable-Derived Coherence State Model for Early Warning and Recovery Profiling of Stress-Related Dysregulation. The study presents a wearable-derived coherence state framework designed to model transitions into a rule-defined dysregulation state, quantify recovery dynamics using a recovery half-life metric, and explore reproducible coherence phenotypes from heart rate variability and sleep-related signals. The analysis uses two open datasets, WESAD (N = 15) and a longitudinal real-world wearable HRV dataset with daily sleep diaries (N = 49). A logistic regression risk model with leave-one-subject-out cross-validation was used to forecast entry into dysregulation, alongside time-to-event modeling, recovery profiling, and unsupervised clustering. Reported results include ROC AUC = 0.645, PR-AUC = 0.550, Brier score = 0.236, and a median early warning lead time of 16.97 hours. This work is retrospective and intended as an interpretable research framework for wearable signals.

This record also includes a supporting ZIP archive containing source and derived analysis files for the real-world wearable cohort used in the study. The archive includes participant survey data, sleep diary data, sensor-derived HRV data, filtered HRV data, derived feature tables, recovery half-life outputs, out-of-fold prediction outputs, summary metadata, and figure image files used in the analysis workflow. The datasets analyzed in the manuscript are publicly available from their original sources, and the ZIP file is provided here as a supporting package for the real-world wearable HRV component of the study.