Evaluating and Improving Fairness in Synthetic Electronic Medical Record Data

The School of EECS is hosting the following PhD Progress Review 1 Confirmation Seminar:

Evaluating and Improving Fairness in Synthetic Electronic Medical Record Data

Speaker: Reji Rony Philip

EECS host: Professor Xue Li

Abstract
Synthetic electronic medical record (EMR) data offer an important opportunity to support healthcare research while reducing barriers associated with privacy, governance and access to real-world clinical data. However, synthetic data cannot be assumed to be fair simply because they preserve the statistical characteristics of the source data. Generative models may reproduce or amplify existing demographic imbalances, potentially affecting subgroup representation and downstream machine-learning performance.

This PhD investigates how fairness in synthetic EMR data can be systematically evaluated and improved. Study 1 comprises a systematic review of fairness in synthetic healthcare data, identifying substantial heterogeneity in fairness definitions, protected attributes, evaluation metrics and generative approaches. Building on these findings, Study 2 will develop a reproducible fairness-evaluation framework using MIMIC-IV and compare representational and algorithmic fairness across synthetic datasets generated using different approaches. Study 3 will investigate whether fairness can be improved by explicitly incorporating fairness objectives or subgroup-aware strategies during synthetic-data generation. Study 4 will evaluate the transferability of the framework to synthetic data generated from real-world Queensland Health EMR data mapped to the OMOP Common Data Model.

The research aims to provide a reusable framework for evaluating and improving fairness while considering associated trade-offs with fidelity, utility and privacy.

Short biography
Reji Rony Philip is a PhD candidate in the School of Electrical Engineering and Computer Science at The University of Queensland. She has professional experience in healthcare data analytics, electronic medical record systems, clinical data extraction and reporting, and digital health research. Her research focuses on fairness in synthetic electronic medical record data, with particular interests in synthetic data generation, machine learning and the OMOP Common Data Model. Her PhD investigates methods for evaluating representational and algorithmic fairness in synthetic healthcare data, improving fairness during synthetic-data generation, and transferring these methods from public benchmark datasets to real-world Australian clinical data.