Continuous Stress Monitoring for Healthcare Workers: Evaluating Generalizability Across Real-World Datasets

Abstract

Healthcare workers are constantly exposed to high-stress working conditions that increase burnout rates and lower the quality of patient care. While it is vital to continuously monitor healthcare worker stress to provide necessary interventions, traditional survey methods can interfere with tasks in real-world healthcare contexts. Wearable devices offer a non-invasive way to detect worker stress continuously, however, predictions may be influenced by context-specific activities and stress semantics. Our work assesses generalized stress detection for health workers with currently available datasets using shared stress representations. We tested four machine learning algorithms (support vector machines, random forest, feed-forward networks, and extreme boosting). We identified extreme boosting as the top-performing model, achieving ROC-AUC scores of up to 0.83 using real-world health worker data. To benchmark generalizability among machine learning stress detection models, we assessed domain adaptation to enhance the transferability of models. Supervised domain adaptation performs comparably to vanilla prediction methods on the target domain. Our work identifies numerous challenges to generalizing stress detection for health workers, including limited shared modalities, small sample sizes, and varying stress definitions. As more data is collected, we must prioritize shared stress representations to enable continuous stress prediction for health workers.