Abstract:
Objective: This study aimed to assess the prevalence and factors associated with the severity of work-related injuries among workers in major industries in Samutprakan province. This study also developed a neural network to forecast lost workdays from occupational injury and tested the efficacy of the regression neural network with the real scenario. Method: A retrospective cross-sectional study was performed in Samutprakarn province by purposive sampling of 1098 work-related injury cases treated in the emergency room of a government hospital between January 2019 and December 2020. Personal and occupational factors were retrieved from the electronic medical records. Severity was defined by days of sick leave in the medical certificate. Personal and occupational factors of workers were analyzed by using the crude odds ratio, 95% confidence interval of OR, and multivariate logistic regression to determine the factors associated with the severity of work-related injuries that require three or more days of sick leave. A total of 1098 cases were divided into 3 categories. First 700 cases as learning set and 300 cases as testing set and the last 98 cases for cross validating which would be the result of network in meeting non-experienced data (network test data). For training network, Python were used and for comparing different network models relatively used RMSE. Result: The prevalence of severe work-related injuries that required three or more days of sick leave in Samutprakan province was 25.4%. The demographic characteristic, based on personal factors, most of them were males (80.6%). The mean age was 35.7±11.8 years, The majority of cases were aged 21-30 years (34.1%) and Thai nationality (55.9%). Work-related injuries often occurred in the day shift (67.7%) and in the construction industry (51.9%). Factors associated with severity of work-related injuries that required three or more days of sick leave with adjusted OR (95% CI) were male 1.8(1.27-2.62), age ≤30 years 1.3(1.01-1.61), foreign workers 1.5(1.15-1.93), afternoon shift 1.32(1.00-1.75) and agriculture and livestock industry 1.88(1.03-3.75), respectively. Finally, the development of neural network model 8 (Hidden layer 1, Node 100) has the momentum learning rule and tanh transfer function which gave the best result of prediction of RMSE at 1.16 and the accuracy was 56.5% Conclusion: The factors associated with severe work-related injuries that required three or more days of sick leave include both personal and occupational factors. These factors should be taken into the occupational safety risk assessment and can be used to create more effective safety measures in the workplace in the future. In this research, the neural network has been utilized for the prediction severity of occupational injury in the number of lost workdays. The best values of RMSE and accuracy of the model were 1.16 and 56.5%. For this purpose, we can use this neural network, model 8 for helping the safety manager for making decisions in the risk estimation and preventive measures of the worker.