The development and validation of two prediction models to identify employees at risk of high sickness absence

C.A.M. Roelen, W. van Rhenen, J.W. Groothoff, J.J. Klink, U. Bultmann, M.W. Heijmans

Research output: Contribution to JournalArticleAcademicpeer-review

Abstract

Background: Sickness absence (SA) is a public health risk marker for morbidity and mortality. The aim of this study was to develop and validate prediction models to identify employees at risk of high SA. Methods: Two prediction models were developed using self-rated health (SRH) and prior SA as predictors. SRH was measured by the categories excellent, good, fair and poor in a convenience sample of 535 hospital employees. Prior SA was retrieved from the employer's register. The predictive performance of the models was assessed by logistic regression analysis with high (>90th percentile) vs. non-high (7lt;90th percentile) SA days and SA episodes as outcome variables and by using bootstrapping techniques to validate the models. Results: The overall performance as reflected in the Nagelkerke's pseudo R2 was 11.7% for the model identifying employees with high SA days and 31.8% for the model identifying employees with high SA episodes. The discriminative ability, represented by the area (AUC) under the receiver operating characteristic (ROC), was 0.729 (95% CI 0.667-0.809) for the model identifying employees with high SA days and 0.831 (95% CI 0.784-0.877) for the model identifying employees with high SA episodes. The Hosmer-Lemeshow test showed acceptable calibration for both models. Conclusions: The prediction models identified employees at risk of high SA, but need further external validation in other settings and working populations before applying them in public and occupational health research and care. © The Author 2012.
Original languageEnglish
Pages (from-to)128-133
JournalEuropean Journal of Public Health
Volume23
Issue number1
DOIs
Publication statusPublished - 2013

Fingerprint

Dive into the research topics of 'The development and validation of two prediction models to identify employees at risk of high sickness absence'. Together they form a unique fingerprint.

Cite this