Post-stroke Discharge Disposition Prediction using Deep Learning

Stroke is the fifth leading cause of death and a major cause of long-term disability in the United States. Because of its life-threatening consequences, the stroke system of care including acute treatment and post-acute rehabilitation is crucial. In order to support the stroke system of care in the age of big data, predictive analytics can be applied to forecast what will happen in the future. For example, we can predict post-stroke discharge disposition at an acute stroke admission, which will facilitate optimizing acute treatment and planning post-acute rehabilitation with the desired outcomes. In this preliminary study, we will explore deep learning for post-stroke discharge disposition prediction and evaluate prediction performance using hospital discharge data provided by Tennessee Department of Health. Deep learning is a powerful tool to deal with large-scale data and provides an end-to-end solution to multi-class classification including dimensionality reduction. Our preliminary results will demonstrate the effectiveness of deep learning and suggest the further exploration for performance improvement and clinical adoption.