Data-driven approach for creating synthetic electronic medical records

Area of Science:

• Computational epidemiology
• Health informatics
• Data science

Background:

• Electronic medical records (EMRs) are crucial for developing disease outbreak detection algorithms.
• Privacy concerns restrict sharing of anonymized EMRs, hindering algorithm validation and comparison.
• A gap exists between advanced bio-surveillance algorithms and the availability of non-identifiable EMR data.

Purpose of the Study:

• To develop a method for generating complete, synthetic electronic medical records (EMRs).
• To enable the comparison of disease outbreak detection algorithms by providing shareable synthetic data.
• To bridge the gap between algorithm development and data accessibility in bio-surveillance.

Main Methods:

• Generation of synthetic patient identities and basic demographic information.
• Identification and adaptation of real patient care patterns for synthetic populations.
• Creation of synthetic EMRs including visit, clinical, laboratory, and radiology data.

Main Results:

• Synthetic EMRs were generated for 203 tularemia outbreak patients and over 3000 background patients (ages 4-11).
• Expert validation identified and corrected errors in 19% of outbreak EMRs and <3% of background EMRs.
• The generated synthetic EMRs provide a viable dataset for algorithm testing.

Conclusions:

• A data-driven methodology for generating fully synthetic EMRs has been successfully developed.
• The method is generalizable to various datasets with similar elements and adaptable to different infectious diseases.
• The approach can be modified to generate synthetic background EMRs for diverse age groups.