Abstract
Purpose Updating the Wallis Occupational Rehabilitation Risk (WORRK) model formula, predicting non-return to work (nRTW) at different time points (3 and 12 months) than in the validation study (2 years). Methods Secondary analysis of two samples was carried out (following orthopaedic trauma), including work status, the first at 3 months (428 patients) and the second at 12 months (431 patients) after discharge from rehabilitation. We used calibration (agreement between predicted probabilities and observed frequencies) and discrimination (area under the receiver operating characteristics curve) to assess performance of the model after fitting it in the new sample, then calculated the probabilities of nRTW based on the coefficients from the 2-year prediction. Finally, the intercepts were updated for both 3- and 12-month prediction models (re-calibration was necessary for the adjustment of these probabilities) and performance re-evaluated. Results Patient characteristics were similar in all samples (mean age 43 in both groups; 86% male at 3 months, 84% male at 12 months). The proportion of nRTW at 3 months was 63.8% and 53.4% at 12 months (50.36% at 2 years). Performance of the original WORRK for both 3- and 12-month prediction showed an AUC of 0.73, while statistically significant miscalibration was found for both time points (p < 0.001). After the updating of the intercept, calibration was improved and did not show significant miscalibration (p = 0.458 and 0.341). The AUC stayed at 0.73. Conclusion The WORRK model was successfully adapted by changing the intercept for 3- and 12-month prediction of nRTW, now available for use in clinical practice.