Development of a Non-Invasive Sleep-Based Prediction Platform for Burnout and Retention Risk Among Postgraduate Nurses: A Psychophysiological and AI-Driven Approach for High-Stress Clinical Populations

Postgraduate nurses frequently experience substantial psychological and physiological stress during the transition from academic training to clinical practice. Heavy workloads, rotating shifts, emotional demands, and adaptation to clinical environments may contribute to sleep disturbance, fatigue, burnout, and increased turnover intention. Previous studies have demonstrated significant associations between sleep quality, autonomic nervous system regulation, emotional distress, and occupational burnout among healthcare workers, particularly in shift-working nurses.

Current psychological assessments mainly rely on self-reported questionnaires and short-term evaluations, which may not adequately capture dynamic physiological changes over time. Recent advances in non-contact sleep monitoring technologies provide opportunities for continuous and low-burden collection of sleep-related physiological data in natural sleep environments. In addition, artificial intelligence and machine learning approaches may improve early identification of individuals at higher risk of burnout and retention problems.

This study is a prospective longitudinal observational study designed to investigate the relationship between sleep-related physiological characteristics, psychological status, burnout risk, and retention risk among postgraduate nurses during the early clinical transition period.

Eligible participants will include newly employed postgraduate nurses within three months of clinical employment. Participants will complete validated psychological questionnaires, including the Brief Symptom Rating Scale-5 (BSRS-5), Chinese Health Questionnaire-12 (CHQ-12), Pittsburgh Sleep Quality Index (PSQI), Karolinska Sleepiness Scale (KSS), and Copenhagen Burnout Inventory (CBI). In parallel, participants will undergo non-invasive and non-contact sleep monitoring under natural sleep conditions. Sleep-related physiological parameters including sleep efficiency, sleep stage distribution, deep sleep proportion, REM sleep stability, heart rate variability, and respiratory variability will be analyzed.

Repeated assessments will be conducted longitudinally at baseline, 3 months, and 6 months. Statistical analyses will include descriptive statistics, longitudinal analyses, generalized estimating equations, mixed-effects models, and survival-related analyses when applicable. Machine learning and deep learning approaches, including Random Forest, XGBoost, and longitudinal prediction models, will be applied to develop predictive models for burnout and retention risk.