The project aims to build a Graph Neural Network to solve various predictive tasks from Electronic Health Records (EHR) in the healthcare domain. Given the complexity and multidimensionality of EHR data, it is crucial to unveil intricate relationships among medical concepts. Thus, we propose an innovative approach employing Graph Neural Networks (GNN) to learn these interconnections in EHR data. Our novel heterogeneous graph network exhibits robustness in understanding graph structures, leading to adaptive performance enhancements across various predictive tasks in EHR. The significance of our work lies in its contribution to healthcare analytics, where the vast data within Electronic Health Records can be leveraged to enhance diagnosis, treatment, and patient care. Utilizing heterogeneous graph neural networks for adaptive multitasking offers the potential for more accurate and insightful applications in the healthcare sector, ultimately benefiting medical professionals and patients.
