Overview
SYNRARE is a graphical user interface (GUI) built upon the Synthea framework, developed for the generation of synthetic Electronic Health Records (EHRs). Its primary function is to create synthetic EHRs for patients with rare diseases, specifically designed to allow for controlled differences from common disease patient populations. This tool serves to facilitate the benchmarking and testing of Machine Learning (ML) algorithms.
Research Context
The diagnosis of rare diseases (RD) frequently experiences delays, often due to symptom similarities with more common disease variants. Machine Learning algorithms, when applied to Electronic Health Records, demonstrate potential for expediting this diagnostic process. However, the utilization of real patient EHRs faces significant obstacles stemming from legal and privacy concerns. Synthetic Data Generation emerges as an alternative approach to obtaining EHRs, enabling their application with ML algorithms for benchmarking and development. Existing synthetic data generation algorithms often lack specific support for creating subsets of patients that exhibit a definable degree of difference from the majority, which is necessary to simulate RD patients.
Approach
SYNRARE was designed as a GUI-based solution within the Synthea framework. This design allows for easier modification and generation of synthetic EHRs. The system is specifically engineered to produce synthetic EHRs for RD patients where the differences from common disease patients are quantifiable and controllable. This capability enables researchers to generate various scenarios for benchmarking and testing their Machine Learning algorithms.
Findings
SYNRARE enables the generation of synthetic EHRs for rare disease patients. These generated EHRs can be configured to differ from EHRs of common disease variants by a pre-defined degree. This functionality supports the benchmarking and testing of Machine Learning algorithms under controlled technical conditions. The system allows researchers to rapidly benchmark their ML algorithms across multiple scenarios.
Availability and Implementation
SYNRARE, along with detailed installation instructions, is publicly available. The resource can be accessed at https://gitlab.sdu.dk/screen4care/synrare.