A Hypergraph-based temporal model for Electron Health Records

Modeling Electronic Health Records is essential for digital medicine. However, Existing approaches do not simultaneously account for higher-order interactions among medical codes and the temporal changes in patient conditions, along with the complex factors in real-world medical scenarios, including model usability. To mitigate these limitations, we develop a new framework referred to as HYTE, which combines hypergraph representation learning with temporal encoding component. This framework introduces accurate inference models, HYTE-T and HYTE-I, tailored for different usage contexts. Experimental evaluations on the EHR dataset underscore the efficacy of the HYTE framework, and comparative experiments highlight the superior performance of hypergraph in representing the co-occurrence among medical codes. Based on the experimental results and analysis, we conclude that HYTE is suitable for application in various real-world scenarios.