Learning Hierarchical Representations of Electronic Health Records for Clinical Outcome Prediction

Clinical outcome prediction based on the Electronic Health Record (EHR) plays\na crucial role in improving the quality of healthcare. Conventional deep\nsequential models fail to capture the rich temporal patterns encoded in the\nlongand irregular clinical event sequences. We make the observation that\nclinical events at a long time scale exhibit strongtemporal patterns, while\nevents within a short time period tend to be disordered co-occurrence. We thus\npropose differentiated mechanisms to model clinical events at different time\nscales. Our model learns hierarchical representationsof event sequences, to\nadaptively distinguish between short-range and long-range events, and\naccurately capture coretemporal dependencies. Experimental results on real\nclinical data show that our model greatly improves over previous\nstate-of-the-art models, achieving AUC scores of 0.94 and 0.90 for predicting\ndeath and ICU admission respectively, Our model also successfully identifies\nimportant events for different clinical outcome prediction tasks\n