Clinical Features for Identifying the Possibility of Toileting Independence after Convalescent Inpatient Rehabilitation in Severe Stroke Patients: A Decision Tree Analysis Based on a Nationwide Japan Rehabilitation Database

Abstract Background and purpose In severe stroke patients, considerable concern should be given to toileting activity in rehabilitative support. Recently, the application of artificial intelligence, including machine learning (ML), has expanded into the stroke medical field, which could clarify the factors affecting toileting independence in severe stroke patients. This study aimed to identify the factors affecting toileting independence in severe stroke patients using ML. Methods We used the Japan Rehabilitation Database from 2005 to 2015 to investigate data from 2292 severe stroke patients. We performed the chi-squared automatic interaction detection (CHAID) algorithm with various explanatory variables. Results The CHAID model identified modified Rankin scale (mRS) score as the first discriminator. Among those with an mRS score ≤4, the next discriminator was age (score ≤72, 73–80, or >80). Among those with an mRS score > 4, the next discriminator was also age (score ≤57, 58–72, 73–80, or >80). Interestingly, some patients achieved toileting independence, although this study focused on severe stroke patients. In branches based on age, the percentage of the patients who achieved toileting independence at discharge decreased progressively with age. Conclusion We identified the influential factors, including reference values, for achieving toileting independence in convalescent severe stroke patients.