NLP Work related stress

Stress in the workplace should be automatically detected and described via routine data in the documentation system.


  • Lead school School of Health Professions
  • Additional schools School of Engineering and Computer Science
  • Institute Nursing
  • Research unit Innovation in the Field of Health Care and Human Resources Development
  • Strategic thematic field Thematic field "Humane Digital Transformation"
  • Funding organisation BFH
  • Duration 01.05.2023 - 30.04.2024
  • Project management Dr. Christoph Golz
  • Head of project Dr. Christoph Golz
  • Project staff Dr. Souhir Ben Souissi


Health organizations struggle with workforce shortages. Nearly 50% quit their job early in their career due to high work-related stress. Until now research on work-related stress predominantly conducted studies by gathering primary data from health professionals with surveys. The response rate within such studies is decreasing, given the already high workload and increase of research projects with surveys. However, continuous monitoring of work-related stress is crucial to elaborate on the effects of taken measures in practice. Thus, research is needed to invest in innovative methods measuring work-related stress without surveying health professionals. Unstructured textual routine data from daily documentation in the electronic health records could be used to identify the stress level on wards using natural language processing. This would allow decision makers to interpret instant live data and take measures to reduce stress.

Course of action

Natural language processing, such as sentiment analysis for detecting stress will be conducted. Primary data on work-related stress from the national STRAIN study and the ongoing third-party funded longitudinal STRAIN 2.0 study will be used as anchor point of the phenomenon work-related stress. In this project we... - conduct a literature search and deep analysis for existing natural language processing approaches, - conduct multiple labeling techniques to and create labeling guidelines to match textual routine data with subjective experienced work-related stress, - collect, annotate and create high-quality datasets from health organizations, which already stated their interest in participating, - design, implement and test our first classifiers, - match and analyse the data, - prepare a report and recommendations for the next steps to proceed with the development of the minimum valuable product in an innovation project.

This project contributes to the following SDGs

  • 9: Industry, innovation and infrastructure