Paper Title
Predict Delayed Discharge with Ensemble Embedding on Electrical Patient Records

Abstract
The Delayed transfer of care (DTOC) remains a severe problem in hospital management. In this research, we explore whether deep methods plus embedding technique would improve the prediction performance of this problem. Specifically, our approach compromises two steps. First, we conduct embedding learning on medical concepts and turn them into vectors. Second, based on the representation vectors, we employ CNN on the prediction of the DTOC. To evaluate the predictive power of our method, we compare different combinations of embedding dimensions and prediction methods. The results show that CNN with two layers at the embedding dimension of 200 gives the best performance regarding the recall rate of 0.791. Compared to other related research, we innovatively synthesized the embedding of both historical data and intra-visit data, and firstly applied the deep learning method to the prediction of delayed transfer of care. Keywords - Embedding, CNN, Delayed Transfer, Deep Learning, Electrical Patient Record