This is a page that curates AI-related papers published worldwide. All content here is summarized using Google Gemini and operated on a non-profit basis. Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.
Generating Multi-Table Time Series EHR from Latent Space with Minimal Preprocessing
Created by
Haebom
Author
Eunbyeol Cho, Jiyoun Kim, Minjae Lee, Sungjin Park, Edward Choi
Outline
RawMed is the first framework to synthesize multi-table time-series EHR data that resembles raw EHRs, unlike traditional medical records that contain only a few key signs or structured codes selected by experts. It captures complex structure and temporal dynamics with minimal preprocessing using text-based representation and compression techniques. We also propose a novel evaluation framework for multi-table time-series synthetic EHRs to evaluate distributional similarity, inter-table relationships, temporal dynamics, and privacy. When validated on two open-source EHR datasets, RawMed outperforms baseline models in terms of fidelity and usability. The code is available at https://github.com/eunbyeol-cho/RawMed .
A novel framework for generating multi-table time series synthetic EHR data similar to raw EHR data is presented.
◦
Efficient data synthesis and capture of complex structures and temporal dynamics using text-based representation and compression techniques.
◦
A novel evaluation framework for multi-table time series synthetic EHR data (distributional similarity, inter-table relationships, temporal dynamics, privacy)
◦
Improved fidelity and usability compared to previous models
◦
Open source code disclosure
•
Limitations:
◦
The paper does not specifically mention Limitations. Additional experiments and verification are needed to clarify generalizability and Limitations.
◦
Since these are validation results for a specific EHR dataset, further research is needed to determine generalizability to other datasets.
