Metabolic risk stratification of night shift workers in a large retail workplace through clustering and SHAP...

While working at night is an important occupational risk factor associated with metabolic diseases, the fact that different workers have different risk patterns has not been fully studied. In this study, we used explainable AI analysis methods to find out and specify what factors contributed to metabolic health risks among night shift workers.

The study was conducted with employees working at night at a large domestic distribution company. Basic information such as health examination data (blood sugar, cholesterol, blood pressure, etc.) and age and working period of the study subjects was collected and analyzed. Using unsupervised learning (UMAP and K-means clustering), people with similar health characteristics were divided into four groups. Using a prediction model called Random Forest and an analysis of Shapley Additive Explanations (SHAP), we found out what health factors had the greatest impact on distinguishing these four groups.

In this study, four health types (clusters) were identified. Cluster 0 is an overall healthy low-risk group, which seems to reflect the "health worker effect." Cluster 1 included workers who were older (average 40.9 years), had longer night work experience (average 13.3 years), and had high blood pressure and cholesterol levels. Cluster 2 is a group of moderate-risk groups with slightly increased body mass index (BMI) and lipid levels, which are interpreted as transitional conditions with deteriorating health. Cluster 3 was found to be young (average 37.4 years), had a short working period (average 6.0 years), and had high BMI, blood sugar, and blood pressure, which led to many irregularities in lifestyle and social jet lag. In addition, because of SHAP analysis, BMI, triglyceride-blood sugar index, blood pressure, and working period were identified as the most important factors in distinguishing these four groups.

It has been confirmed that the metabolic health risks of night workers vary from person to person, and that these differences can be effectively classified (tiered) through machine learning analysis. The results of this study serve as an important foundation for developing customized industrial health strategies, such as redesigning work schedules or personalized health care and prevention programs.