Medical records generated during occupational health surveillance processes have large amounts of unexploited information that can help to reduce silica-related health risks and many occupational diseases. The methodology applied in this study consists in analyzing through machine learning techniques a database with 70,000 medical examinations from workers in the energy and construction industry in Spain. First, a general unsupervised Bayesian model is built and node force analysis is used to identify the factors with the greatest impact on the worker's health surveillance process. Second, a predictive Bayesian model is created and mutual information is employed to assess the more relevant factors affecting the medical capability of workers exposed to silica dust. The lung auscultation and the breathing exploration are the two factors that influence the most the medical capability of silica-exposed employees. Probabilistic inference shows a remarkable gender effect, where women present more resilience towards occupational diseases than men showing a higher proportion of normal results in certain key factors, such as body mass index (♀49.73%, ♂25.17%) or spirometry (♀53.73%, ♂48.91%). Finally, environmental conditions demonstrate to have a major influence on spatial variability of occupational diseases. The design of health prevention programs based on geographical variations can be crucial to the attainment of an ongoing and sustained healthier workforce with a reduction in the number of chronic workplace illnesses.
Keywords: Environment; Health risk; Machine learning; Occupational health; Silica; Spatial variability.