Machine learning to predict return to work after medical rehabilitation for musculoskeletal disorders

Purpose: This study aimed to identify and quantify the contribution of various features and their interactions to the prediction of return to work (RTW) following medical rehabilitation due to musculoskeletal disorders, using a machine learning approach. In addition, the study explored whether the COVID-19 pandemic affected RTW outcomes or moderated the effects of established predictors. Methods: A Light Gradient Boosting Machine (LightGBM) model was trained and validated using data from the Rehabilitation Statistics Database (RSD) of the German Pension Insurance. Model predictions were explained using Shapley values to determine the contributions of individual features and their interactions. A total of 685,890 individuals who received multimodal medical rehabilitation for musculoskeletal disorders via the German Pension Insurance between January 2018 and December 2021 were included in the study. Results: The model demonstrated high predictive performance (Accuracy: 0.830; Precision: 0.842; Recall: 0.908; AUC: 0.882), exceeding that of logistic regression models reported in the literature. The five most important features and interactions for predicting RTW were: (1) the number of days in employment in the calendar year prior to rehabilitation, (2) the duration of incapacity for work in the twelve months prior to rehabilitation, (3) employment income in the calendar year prior to rehabilitation, (4) work capacity at discharge, and (5) the interaction between the duration of incapacity for work and employment days. The COVID-19 pandemic cohort variable and its interactions were not among the top 15 contributors. Conclusion: The use of machine learning proved valuable for predicting RTW after medical rehabilitation. The approach confirmed key predictors identified in previous research, such as prior employment and duration of incapacity for work. In addition, the model identified meaningful interactions between features and highlighted several previously underexplored factors, such as work capacity at discharge, that contribute to RTW outcomes. Regarding the COVID-19 pandemic, the results align with earlier findings, suggesting that the pandemic had only a minor influence on RTW.