Predictive models for cognitive rehabilitation of patients with traumatic brain injury

Traumatic Brain Injury (TBI) is a leading cause of disability worldwide. Computerized rehabilitation tasks are increasingly replacing traditional paper and pencil approaches in cognitive rehabilitation treatments of TBI patients in clinical practice. Therapists usually decide treatment configuration (TC) (e.g. total number of sessions or tasks per patient) based on intuition. Predictive techniques have traditionally been applied for cognitive rehabilitation gross outcome prognosis (e.g. cognitive improvement or not), without considering TC variables. In this work we propose to enrich predictive models with variables that therapists can act upon. We statistically compared 48 predictive techniques (with extensive parameters tuning), from 12 predictive models considering 3 different resampling methods with and without TC variables. We applied model-dependent and model-independent ranking techniques to assess variables' importance. We analyzed the contribution of TC variables for prediction of response to treatment of 415 severe TBI patients that performed 148710 cognitive rehabilitation tasks. We identified predictive models and techniques with TC variables outperforming those without TC variables. We obtained superior performance (72.7% accuracy) to previous state-of-the-art models with similar datasets. We found highly ranked TC variables after importance evaluation analysis. Finally we suggest use cases including the obtained predictive models that contribute to treatments personalization and efficiency.