Development and validation of machine learning prediction model for post-rehabilitation functional outcome after intracerebral hemorrhage

Objective: Predicting outcomes after intracerebral hemorrhage (ICH) may help improve patient outcomes. We developed and validated a machine learning prediction model for post-rehabilitation functional outcomes after ICH. Patient selection and explanatory variable settings were based on clinical significance. Functional outcomes were predicted using ternary classification. Methods: The subjects were patients aged > 18 years without pre-onset severe disability who developed primary putaminal and/or thalamic hemorrhage and underwent an inpatient rehabilitation program. As explanatory variables, 43 values related to patient background, imaging-related findings, systemic conditions, neurological findings, and blood tests were acquired within 10 days of onset. As an objective variable, the functional outcome at discharge to home or nursing home was acquired using a ternary classification. The dataset consisting of the collected information was split into a training dataset and a test dataset with a ratio of 2:1. A predictive model using a balanced random forest algorithm was created using supervised learning from the training dataset. The predictive performance was validated using a test dataset. Results: Between January 2018 and June 2019, 100 consecutive patients were included in the study. The areas under the receiver operating characteristic curves for predictions of good, moderate, and poor outcomes were 0.952, 0.790, and 0.921, respectively. Conclusions: The predictive performance of the model was comparable to that of previous models. Patient selection and variable settings from a clinical perspective may contribute to accurate and detailed predictions. These study designs are based on design thinking and may meet the needs of clinical practice.