Influencing factors of fall injury risks in manufacturing enterprises based on Bayesian networks
Falls are one of the most important types of occupational injuries. The incidence of falls is high in manufacturing workers. However, most of the studies on falls in China focus on primary and secondary school students and the elderly, and there are few studies on falls in the occupational population.
To evaluate efficiency of Bayesian network model in predicting fall injury risks in manufacturing enterprise staff, and impacts from work content, work environment, enterprise status, and health management on falls and their mutual relationships, and provide a scientific basis for enterprises to carry out fall-associated injury intervention.
Data from the European Survey of Enterprises on New and Emerging Risks (ESENER) were used. The survey provided data on work content, working environment, enterprise status, and health management of enterprises in European countries. The outcome indicator, was fall injury risks reported in enterprises. A total of 23 potential impact factors covering work content, working environment, enterprise status, and health management were screened by least absolute shrinkage and selection operator (LASSO) regression, followed by Bayesian network model for structure learning and parameter learning and area under the curve (AUC) for model fitness evaluation, using R and Netica 5.18. Diagnostic inference analysis was also conducted to identify key influencing factors and key influencing chains of fall injury risks based on the change rate of fall injury risks.
In 5997 enterprises surveyed, 2573 (42.9%) enterprises reported fall injury risks. Ordered by their coefficient estimates from high to low, the 14 variables (mean-squared error=0.20) selected by LASSO regression were: manual handling, repetitive arm movement, poor posture, using desktop computers, and using robots in the category of work content; abnormal temperature and noise in the category of working environment; company size and employee quality in the category of enterprise status; mental health training, regular risk assessment, availability of psychologists, health and safety procedures, and provision of psychological counseling in the category of health management. The fitting result of Bayesian network model for fall injury risks was good (AUC=0.779). The Bayesian network diagnostic inference identified five key influencing factors, including abnormal temperature (change rate=35.9%), poor posture (change rate=27.3%), noise (change rate=23.4%), manual handling (change rate=18.2%), and repetitive arm movement (change rate=5.1%). The key influencing chain was "manual handling - poor posture - repetitive arm movement - fall injury risks" (combined change rate=16.9%).
The Bayesian network model has a good predictive performance in predicting the risk of falls in manufacturing enterprises. Manufacturing enterprises need to focus on jobs involving manual handling and repetitive arm movement, identify and improve workers' poor posture and mental health problems, and avoid workers working in harsh temperature or noise environment.