Which Return-to-Work Expectation Question is Most Accurate for Predicting Prolonged Sick Leave? A Secondary Analysis from a Randomised Controlled Trial

Abstract Purpose To compare the prognostic value of four distinct return-to-work (RTW) expectation questions for predicting prolonged sickness absence. Methods Secondary analysis of a randomised clinical trial using baseline data from 675 workers sick-listed for 8 weeks, with 12-month registry follow-up. Four RTW expectation questions were evaluated for the primary outcome of sick leave exceeding 180 days, and secondary outcomes of sick leave exceeding 90 days and disability benefits. Logistic regression was used to estimate associations in simple and fully adjusted models. Incremental prognostic value was assessed using change in discrimination (Δc-statistic) and explained variance (ΔR 2 ). Clinical utility was evaluated using decision curve analysis, and category-free net reclassification improvement (NRI) was calculated for each predictor. Results Long-term sick leave exceeding 180 days occurred in 57% of participants. All four expectation predictors were independently associated with all three outcomes. Adding expectation predictors to a simple adjusted model yielded substantial gains in discrimination for the primary outcome (Δc 0.05–0.14), whilst incremental gains were modest in the fully adjusted model that included work ability (Δc 0.01–0.02). Decision curve analysis confirmed clinical utility primarily in low-information contexts. Questions incorporating a specific time frame outperformed those without. NRI gains were driven by improved classification of non-events. Conclusion Questions asking sick-listed individuals about their RTW expectations are robust predictors of actual RTW regardless of how the question is asked. Expectation questions including a time frame should be preferred as they performed better than those without a time frame. Trial Registration The data derive from a randomised controlled trial prospectively registered at ClinicalTrials.gov (NCT03212118, registered 11 July 2017).