A nurse-centered edge-server wearable system for real-time monitoring of occupational stress in clinical care.

Background and objectives: Occupational stress among clinical nurses is a long-standing quality-of-care concern associated with burnout, attrition, lapses in patient safety, and erosion of workforce wellbeing. Continuous, nurse-centered monitoring of within-shift physiological stress states is therefore an emerging priority for nursing innovation and quality improvement (QI), yet routine clinical adoption of wearable monitoring is constrained by accuracy, response time, communication load, alert fatigue, and physiological differences across individual nurses. This study proposes the Nurse-centered Edge-Server Technology (NEST), a wearable edge-server framework for real-time occupational-stress monitoring designed to support QI-oriented screening and workflow awareness. Methods: Physiological and motion signals from wearable devices are processed locally with a compact feature set for rapid screening, while a richer feature set confirms suspected high-stress events through cascade alerting. A central server collects brief summaries from each device and returns small updates that personalize the model to each nurse over time. The framework was evaluated on the public Wearable Stress and Affect Detection (WESAD) dataset and on a continuous nurse-monitoring dataset collected in clinical wards, using subject-wise evaluation, out-of-fold supplementary event analysis, ablation experiments, and edge-equivalent runtime tests. Results: Lightweight screening preserved accuracy comparable to that of a richer feature set while reducing the feature dimension from 48D to 15D. In a held-out event-level test set of 24 positive events, the selected cascade preserved the 14/24 event hit count observed with Single48 while reducing false alerts from 9.68 to 1.88 per hour and total alerts from 10.65 to 2.10 per hour. Confidence intervals remained wide because the strict held-out stream protocol yielded 24 positive test events; supplementary all-subject and multi-split analyses were used as robustness checks for the primary held-out evaluation. In the edge-equivalent software setting, P95 latency remained in the millisecond range, the P95 CPU duty cycle was 6.77%, and 15D feature-summary upload was 4.69% of raw-stream upload. Chronological personalized updating increased sensitivity when feedback updates were available, but it did not improve overall F1 and lowered specificity; no-update controls showed no artificial gain. Conclusions: The retrospective evaluation supports the technical feasibility and alert-control value of a wearable edge-server pipeline that combines compact inference, cascade event confirmation, and personalized updating as an assistive QI-oriented monitoring approach. The study evaluates prediction behavior, alert burden, latency, and communication load; direct effects on burnout, nurse retention, workload reduction, patient-safety outcomes, or occupational-health improvement require prospective workflow studies. NEST is intended to augment clinical judgement and the relational role of nurses in occupational-health management and workforce sustainability.