Prediction of Work-relatedness of Shoulder Musculoskeletal Disorders as by Using Machine Learning

Background

This study aimed to develop prediction models for the work-relatedness of shoulder diseases through machine learning algorithms.

Methods

The dataset comprised 7,270 cases of 8,302 individuals who applied for occupational diseases and received the final approval decision from the Korea Workers' Compensation and Welfare Service's Disease Evaluation Committee, which is related to shoulder musculoskeletal disorders between January 2020 and December 2021. In this study, demographic analysis and difference of approval rate by shoulder diseases were performed. Additionally, machine learning algorithms, including logistic regression, support vector machine, decision tree, random forest, and the XGBoost, were utilized to construct prediction models for work-relatedness assessment.

Results

The performance of each model was evaluated. XGBoost showed an accuracy of 81.64% and an area under the curve of 0.73, and random forest showed an accuracy of 84.46% and an area under the curve of 0.73. Key factors influencing work-relatedness assessment were employment period, physical burden score, gender, and age.

Conclusion

The application of various machine learning techniques showed high performance score, representing that it would be helpful to reduce the differences in judgment between occupational environment medicine physicians.