Using machine learning to investigate earning capacity in patients undergoing psychosomatic rehabilitation—A retrospective health data analysis

Psychiatric disorders increasingly contribute to disability and early retirement. This study was conducted to investigate whether machine learning can contribute to a better understanding and assessment of such a reduced earning capacity. It analyzed whether impaired earning capacity is reflected in missing treatment effects, and which interventions drive treatment effects during psychosomatic rehabilitation. Analyses were based on routine clinical data encompassing demographics, diagnoses, psychological questionnaires before, and after treatment, interventions, and an interdisciplinary assessment of earning capacity for N = 1,054 patients undergoing psychosomatic rehabilitation in 2019. Classification of patients by changes in self-reported mental health and interventions predictive of changes were analyzed by gradient boosted model. Clustering results revealed three major groups, one of which was comprised almost exclusively of patients with full earning capacity, one of patients with reduced or lost earning capacity and a third group with mixed assessments. Classification results (Kappa = 0.22) indicated that patients experienced modestly divergent changes over the course of rehabilitation. Relative variable influence in the best model was highest for changes in psychological wellbeing (HEALTH-49). Regression analysis identified intervention A620 (physical exercise therapy with psychological goal setting) as most influential variable predicting changes in psychological wellbeing with a model fit of R(2) = 0.05 (SD = 0.007). Results suggest that disability due to psychiatric disorders does associate with distinct demographic and clinical characteristics but may be less clear-cut in a subgroup of patients. Trajectories of treatment response show moderately divergent paths between patient groups. Moreover, results support both physical exercise therapy as efficient intervention in reducing disability-associated impairments and the complementarity of a multimodal treatment plan.