The study introduces ‘WellFactor’, a method that creates comprehensive patient profiles using not only traditional medical records but also diverse data sets from healthcare web portals. The approach uses constrained low-rank approximation to handle the sparsity often found in healthcare data. It refines embedding results by incorporating task-specific label information, providing a more informed perspective on patients. It can instantaneously compute embeddings for new patient data, eliminating the need to revisit the entire data set or recomputing the embedding. Evaluations on real-world healthcare data show that WellFactor outperforms existing methods in classification performance, meaningful patient clustering, and consistent patient similarity searches and predictions.
Publication date: 22 Dec 2023
Project Page: Not provided
Paper: https://arxiv.org/pdf/2312.14129
