A Multimodal Physiological Dataset for Nurse Stress Detection

The daily life of a nurse is a constant negotiation between human needs and rigid schedules. They care for patients while managing high workloads, emotional demands, and the physical toll of long shifts. For decades, researchers have known that this environment creates significant stress, which can harm the nurse's health and the quality of patient care. Traditionally, understanding this stress has relied on asking people how they feel, often long after the fact, or looking at broad statistics about burnout. However, stress is not a single thing that can be measured with one tool; it is a complex reaction involving how a person thinks about their day, how their body physically responds in the moment, and the specific circumstances of their work schedule. The challenge has been finding a way to see all these pieces together in real time, without disrupting the very work being studied.

A team of researchers in Switzerland set out to build a new kind of picture of nurse stress by combining three different sources of information: what nurses said they felt, what their bodies were doing, and what their work schedules actually looked like. They recruited twenty-four nurses working in a hospital and asked them to wear a smartwatch for three months. These watches continuously recorded the nurses' heart rates and body temperatures while they worked. After every shift, the nurses filled out a short digital questionnaire about their satisfaction with the day, how stressed they felt, and how well the shift fit with their personal lives. The researchers also collected the hospital's official records of the nurses' shifts, including both the planned schedule and the actual schedule that was worked, which allowed them to see where changes or disruptions occurred.

The goal was to see if these different streams of data could be woven together to identify when a nurse was under stress. The researchers first had to turn the daily questionnaire answers into a clear label for each shift, categorizing them as not stressed, mildly stressed, or very stressed based on a combination of the answers. They then matched these labels with the physiological data from the watches and the details of the work schedule. The result was a massive dataset containing over sixteen thousand individual observations, each linking a specific moment in time to a heart rate, a temperature, a shift type, and a stress label. This collection of data represents a rare glimpse into the real-world working conditions of nurses, moving beyond the controlled environments of a laboratory to the busy, unpredictable reality of a hospital floor.

When the team tested whether a computer program could learn to spot stress using this data, they found that looking at the body alone was not enough. A model that tried to predict stress using only heart rate and body temperature performed poorly, often guessing wrong. The situation changed dramatically when the researchers added the context of the work. The computer program became much more accurate when it was allowed to consider the specific shift the nurse was working, the day of the week, and even the identity of the nurse themselves. The best-performing model, which combined the physiological signals with the schedule details and the nurse's identity, correctly identified stress levels about seventy-three percent of the time. This suggests that to understand stress in this profession, you cannot simply look at a beating heart; you must understand the context in which that heart is beating.

The study also shed light on the nature of the work itself. The researchers analyzed the schedules to see if certain patterns, like working many nights in a row or having a day off that suddenly turned into a workday, were linked to higher stress. They found that while some patterns, such as having more free days, were associated with less conflict between work and private life, the relationship was not simple. For instance, a mismatch between a planned schedule and the actual schedule did not always mean the nurse was stressed; sometimes a change was voluntary or welcomed. This highlights that the raw numbers of a schedule do not tell the whole story; the meaning of a shift depends on the individual's circumstances and choices.

Throughout the three-month period, the nurses reported that using the smartwatch was generally easy and acceptable, with an average usability score indicating that the technology did not interfere significantly with their work. This is a crucial finding, as it suggests that collecting this kind of detailed, continuous data in a busy hospital is feasible without overwhelming the staff. However, the researchers were careful to note that their findings are a starting point rather than a final solution. The group of nurses was relatively small, and the models performed best when they could "remember" specific individuals, which means the system learned more about those particular people than about stress in general. The study did not prove that a specific schedule causes stress, but it did demonstrate that combining what people feel, what their bodies do, and what their schedules look like provides a much clearer picture than any single piece of information could offer alone.

Ultimately, this work shows that the path to understanding and managing occupational stress in healthcare lies in integration. By bringing together the subjective experience of the worker, the objective signals of their body, and the structural reality of their schedule, researchers can begin to build tools that are sensitive to the complex reality of nursing. The data collected in this study is now available for others to use, offering a foundation for developing better ways to support the well-being of those who care for the sick. The work confirms that while stress is a personal and internal experience, it is deeply rooted in the external world of shifts, schedules, and the daily rhythm of the hospital.