MD-Manifold: A Medical Distance Based Manifold Learning Approach for Heart Failure Readmission Prediction

Dimension reduction is considered as a necessary technique in Electronic Healthcare Records (EHR) data processing. However, no existing work addresses both of the two points: 1) generating low-dimensional representations for each patient visit; and 2) taking advantage well-organized medical concept structure domain knowledge. Hence, we propose new framework to generate records by combining concept-structure based distance with manifold learning. To demonstrate efficacy, generated hospital visits heart failure patients, which was further used 30-day readmission prediction. The experiments showed great potential proposed (AUC = 60.7%) that has comparative predictive power state-of-the-art methods, including one hot encoding 60.1%) PCA 58.3%), much less training time (improved 99%). can also be generalized various healthcare-related prediction tasks, such mortality