EHR2Trace: Open Source Auditable EHR Data Infrastructure for Clinical AI
EHR2Trace is a new system to standardize and audit electronic health record (EHR) data, enabling reliable model training and evaluation for patient world models and clinical agents. It provides traceable event conversion, source linkage, and export for OMOP and MEDS with automated validation.
EHR2Trace is a tool that converts EHR data from various sources into standardized, traceable patient event representations. Its design enables consistent model training and evaluation for machine learning applications in clinical settings. The system addresses inconsistencies in EHR records by linking each patient event to its source, distinguishing between event occurrence and information availability, and clearly separating medication orders, dispensing, and administration.
Key Capabilities
- Converts heterogeneous EHR data into auditable, traceable events for modeling.
- Maintains links to underlying source records for transparency.
- Separates event time from when information becomes available to downstream systems.
- Distinctly models medication orders, dispensing, and administration.
- Exports to OMOP and MEDS formats with automated validation and reproducible builds.
- Automatically detects data faults during conversion.
EHR2Trace Validation Results
- Total events converted
- 846.4 million (across 3 clinical datasets)
- Automatic validation
- All applicable checks passed except one unit-consistency check on MIMIC-IV
- Fault detection
