Predicting long-term sickness absence among employees with frequent sickness absence

Purpose: Frequent absentees are at risk of long-term sickness absence (SA). The aim of the study is to develop prediction models for long-term SA among frequent absentees. Methods: Data were obtained from 53,833 workers who participated in occupational health surveys in the period 2010–2013; 4204 of them were frequent absentees (i.e., employees with ≥ 3 SA spells in the year prior to the survey). The survey data of the frequent absentees were used to develop two prediction models: model 1 including job demands and job resources and model 2 including burnout and work engagement. Discrimination between frequent absentees with and without long-term SA during follow-up was assessed with the area under the receiver operating characteristic curve (AUC); (AUC) ≥ 0.75 was considered useful for practice. Results: A total of 3563 employees had complete data for analyses and 685 (19%) of them had long-term SA during 1-year follow-up. The final model 1 included age, gender, education, marital status, prior long-term SA, work pace, role clarity and learning opportunities. Discrimination between frequent absentees with and without long-term SA was significant (AUC 0.623; 95% CI 0.601–0.646), but not useful for practice. Model 2 showed comparable discrimination (AUC 0.624; 95% CI 0.596–0.651) with age, gender, education, marital status, prior long-term SA, burnout and work engagement as predictor variables. Differentiating by gender or sickness absence cause did not result in better discrimination. Conclusions: Both prediction models discriminated significantly between frequent absentees with and without long-term SA during 1-year follow-up, but have to be further developed for use in healthcare practice.