Blending Knowledge in Deep Recurrent Networks for Adverse Event\n Prediction at Hospital Discharge

Abstract

Deep learning architectures have an extremely high-capacity for modeling\ncomplex data in a wide variety of domains. However, these architectures have\nbeen limited in their ability to support complex prediction problems using\ninsurance claims data, such as readmission at 30 days, mainly due to data\nsparsity issue. Consequently, classical machine learning methods, especially\nthose that embed domain knowledge in handcrafted features, are often on par\nwith, and sometimes outperform, deep learning approaches. In this paper, we\nillustrate how the potential of deep learning can be achieved by blending\ndomain knowledge within deep learning architectures to predict adverse events\nat hospital discharge, including readmissions. More specifically, we introduce\na learning architecture that fuses a representation of patient data computed by\na self-attention based recurrent neural network, with clinically relevant\nfeatures. We conduct extensive experiments on a large claims dataset and show\nthat the blended method outperforms the standard machine learning approaches.\n

Ähnliche Arbeiten