GRASP: Generic Framework for Health Status Representation Learning Based on Incorporating Knowledge from Similar Patients

Deep learning models have been applied to many healthcare tasks based on electronic medical records (EMR) data and shown substantial performance. Existing methods commonly embed the of a single patient into representation for tasks. Such learn inadequate representations lead inferior performance, especially when patient’s is sparse or low-quality. Aiming at above problem, we propose GRASP, generic framework models. For given patient, GRASP first finds patients in dataset who similar conditions results (i.e., patients), then enhances prognosis by leveraging knowledge extracted from these patients. defines similarities with different meanings between clinical tasks, useful information accordingly, learns cohort extract valuable contained The fused current conduct final Experimental evaluations two real-world datasets show that can be seamlessly integrated state-of-the-art consistent performance improvements. Besides, under guidance experts, verified findings are existing knowledge, indicating generate insights relevant predictions.