Tom Duchemin1, Avner Bar-Hen, Radowan Lounissi, William Dab, Mounia N Hocine. 1. MESuRS laboratory (Modelling, Epidemiology and Surveillance of Health Risks), Conservatoire National des Arts et Métiers (National Conservatory of Arts and Crafts) (Mr Duchemin, Dr Dab, Dr Hocine); Malakoff médéric humanis, (Mr Duchemin, Ms Lounissi); CEDRIC laboratory (Centre for Studies and Research in Computer Science), Conservatoire National des Arts et Métiers (National Conservatory of Arts and Crafts) (Dr Bar-Hen), Paris, France.
Abstract
OBJECTIVE: We hierarchized a range of individual and occupational factors impacting the occurrence of very short (1-3 days), short (4 days to 1 month), or long-term (more than a month) sick leave spells. METHODS: Data were collected from a repeated cross-sectional survey conducted in the French private sector over the period 2011 to 2017. Fifty one sick leave determinants were ranked using a conditional random forest approach. RESULTS: The main determinants of long-term sick leaves were mainly health-related characteristics, such as perceived health, but also work-related covariates such as supervisor acknowledgment. On the contrary, very short-term spells were mainly defined by sociodemographic covariates. CONCLUSION: These results could be useful for devising appropriate actions to prevent against sick leave at the workplace, particularly long-term spells. Random forest approach is a promising approach for ranking correlated covariates from large datasets.
OBJECTIVE: We hierarchized a range of individual and occupational factors impacting the occurrence of very short (1-3 days), short (4 days to 1 month), or long-term (more than a month) sick leave spells. METHODS: Data were collected from a repeated cross-sectional survey conducted in the French private sector over the period 2011 to 2017. Fifty one sick leave determinants were ranked using a conditional random forest approach. RESULTS: The main determinants of long-term sick leaves were mainly health-related characteristics, such as perceived health, but also work-related covariates such as supervisor acknowledgment. On the contrary, very short-term spells were mainly defined by sociodemographic covariates. CONCLUSION: These results could be useful for devising appropriate actions to prevent against sick leave at the workplace, particularly long-term spells. Random forest approach is a promising approach for ranking correlated covariates from large datasets.