Modeling sickness absence data: A scoping review

The identification of sick leave determinants could positively influence decision making to improve worker quality of life and to reduce consequently costs for society. Sick leave is a research topic of interest in economics, psychology, health and social behaviour. The question of choosing an appropriate statistical tool to analyse sick leave data can be challenging. In fact, sick leave data have a complex structure, characterized by two dimensions: frequency and duration, and involve numerous features related to individual and environmental factors. We conducted a scoping review to characterize statistical approaches to analyse individual sick leave data in order to synthesise key insights from the extensive literature, as well as to identify gaps in research. We followed the PRISMA methodology for scoping reviews and searched Medline, World of Science, Science Direct, Psycinfo and EconLit for publications using statistical modeling for explaining or predicting sick leave at the individual level. We selected 469 articles from the 5983 retrieved, dated from 1981 to 2019. In total, three types of model were identified: univariate outcome modeling using for the most part count models (438 articles), bivariate outcome modeling (14 articles), such as multistate models and structural equation modeling (22 articles). The review shows that there was a lack of evaluation of the models as predictive accuracy was only evaluated in 18 articles and the explanatory accuracy in 43 articles. Further research based on joint models could bring more insights on sick leave spells, considering both their frequency and duration.