Background: Stress-related disorders are leading causes of long-term sickness absence (SA) and there is a great need for decision support tools to identify patients with a high risk for long-term SA due to them.
Aims: To develop a clinically implementable prediction model for the duration of SA due to stress-related disorders.
Methods: All new SA spells with F43 diagnosis code lasting >14 days and initiated between 2010-01-01 and 2012-06-30 were identified through data from the Social Insurance Agency. Information on baseline predictors was linked on individual level from other nationwide registers. Piecewise-constant hazard regression was used to predict the duration of the SA. Split-sample validation was used to develop and validate the model, and c-statistics and calibration plots to evaluate it.
Results: Overall 83,443 SA spells, belonging to 77,173 individuals were identified. The median SA duration was 55 days (10% were >365 days). Age, sex, geographical region, employment status, educational level, extent of SA at start and SA days, outpatient healthcare visits, and multi-morbidity in the preceding 365 days were selected to the final model. The model was well calibrated. The overall c-statistics was 0.54 (95% confidence intervals: 0.53-0.54) and 0.70 (95% confidence intervals: 0.69-0.71) for predicting SA spells >365 days.
Limitations: The heterogeneity of the F43-diagnosis and the exclusive use of register-based predictors limited our possibility to increase the discriminatory accuracy of the prediction.
Conclusion: The final model could be implementable in clinical settings to predict duration of SA due to stress-related disorders and could satisfyingly discriminate long-term SA.
Keywords: Duration; Long-term sickness absence; Prediction; Sick leave; Stress-related disorders.
Copyright © 2019 Elsevier B.V. All rights reserved.