Generating Multi-Table Time Series EHR from Latent Space with Minimal Preprocessing

Researchers have developed RawMed, the first framework to generate synthetic multi-table time-series Electronic Health Records (EHR) that closely resembles raw medical data. The system addresses privacy concerns in healthcare data sharing while maintaining fidelity and utility, outperforming baseline models in validation tests.

• →RawMed is the first framework capable of synthesizing multi-table, time-series EHR data that resembles raw medical records.
• →The system uses text-based representation and compression techniques to capture complex structures with minimal preprocessing.
• →A new evaluation framework was developed to assess distributional similarity, inter-table relationships, temporal dynamics, and privacy.
• →Validation on two open-source EHR datasets shows RawMed outperforms existing baseline models in fidelity and utility.
• →The framework addresses critical privacy concerns in healthcare data sharing while enabling research applications.