WellFactor: Patient Profiling using Integrative Embedding of Healthcare Data

• DOI: 10.48550/arxiv.2312.14129
• Publication Date: 2023-01-01
• Authors: Dongjin Choi; Andy Xiang; Özgür Öztürk; Deep Shrestha; Barry Drake; Hamid Haidarian; Faizan Javed; Haesun Park
• Keywords: Profiling (computer programming); Similarity (geometry); Data set

In the rapidly evolving healthcare industry, platforms now have access to not only traditional medical records, but also diverse data sets encompassing various patient interactions, such as those from web portals. To address this rich diversity of data, we introduce WellFactor: a method that derives profiles by integrating information these sources. Central our approach is utilization constrained low-rank approximation. WellFactor optimized handle sparsity often inherent in data. Moreover, incorporating task-specific label information, refines embedding results, offering more informed perspective on patients. One important feature its ability compute embeddings for new, previously unobserved instantaneously, eliminating need revisit entire set or recomputing embedding. Comprehensive evaluations real-world demonstrate WellFactor's effectiveness. It produces better results compared other existing methods classification performance, yields meaningful clustering patients, and delivers consistent similarity searches predictions.