TransLSTD: Augmenting hierarchical disease risk prediction model with time and context awareness via disease clustering

Tao You, Qiaodong Dang, Qing Li, Peng Zhang, Guanzhong Wu, Wei Huang

Abstract: Highlights•A hierarchical model (TranLSTD) is proposed for disease risk prediction, which cannot only extract the developing patterns of patients’ health status in a fine-grained manner, but also provide interpretability of the model at two level.•With the design of classifying disease types based on their duration, a novel type-aware self-attention mechanism is proposed to distinguish different types of diseases.•Substantial experiments on two real-world datasets demonstrate the outstanding performance of the proposed TransLSTD compared to the other state-of-the-art baseline models in terms of accuracy and interpretability.