JMIR Publications

ABSTRACT

Background:

Work-related fatigue has been a common problem in the workplaces. The healthcare providers are particularly prone to work-related fatigue, which may jeopardize their own health and put patient safety at risk. Past studies often conducted in a subjective way by using self-report measures completed after work. With the advancement of wearable technology and machine learning (ML), it is possible to utilize the real-time sensor data as the features to build a prediction model of work-related fatigue for clinical staff.

Objective:

We sought to apply the ML approaches based on data objectively collected from the smartwatches to construct the prediction models of work-related fatigue for emergency healthcare providers.

Methods:

This prospective observational study was conducted at the emergency department (ED) of a tertiary teaching hospital from Mar. 10 to Jun. 20, 2021. The ED physicians, nurses, and nurse practitioners were recruited for this study. All participants wore a commercially available smartwatch capable of measuring various physiological data by the embedded sensors during the experiment. Data were synchronized to a private cloud by the smartphone-constructed gateways via Bluetooth protocol. Participants completed the Multidimensional Fatigue Inventory (MFI) web form before and after each of their work shift. The score differences between each pair of work shifts were calculated and labeled. Several tree-based algorithms were used to construct the prediction models of work-related fatigue based on features collected from the smartwatch. The selected features were ranked by using the entropy measures based on information gain theory. Records were split into training/validation and testing sets at a 70-to-30 ratio and the performances were evaluated using area under the curve (AUC) measure of receiver operating characteristic on the test set.

Results:

In total, 110 participants included in this study, contributing to a set of 1542 effective records. Of them, 85 (5.5%) records were labeled as having work-related fatigue when setting the in-between MFI difference of two standard deviations as the threshold. The mean age of the participants was 29.6. Most of the records were collected from nurses (87.7%) and female (77.5%). A union of 31 features were selected to construct the models. For total participants, CatBoost classifier achieved the best performances of AUC (0.838, 95% CI: 0.742 – 0.918) to identify work-related fatigue. By focusing on a subgroup of nurses under 35 years, eXtreme Gradient Boosting (XGBoost) classifier obtained excellent performance of AUC (0.928, 95% CI: 0.839 – 0.991) on the test set in this study.

Conclusions:

By using features derived from a smartwatch, we successfully built the ML models capable of classifying the risk of work-related fatigue in the ED environment. It is possible to apply such smartwatch-based ML models to predict work-related fatigue and adopt preventive measures for emergency healthcare providers in the future. Clinical Trial: None