LLM-Based Synthetic Tabular Data Generation for Health Equity

Poster Number: P60
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Machine Learning, Fairness and Elimination of Bias, Health Equity, Clinical Decision Support
Primary Track: Foundations
Programmatic Theme: Clinical Informatics

Disparities in model performance for marginalized groups in healthcare prediction tasks are often due to lower representation of these groups in health datasets. We leverage OpenAI’s GPT4 model to generate synthetic data specific to smaller subgroups. Augmenting health datasets with this synthetic data can improve model performance for smaller groups, even compared to other augmentation and reweighting baselines. LLMs hold great promise in synthetic data generation, particularly in sparse-data environments arising from health equity challenges.

Speaker(s):
Daniel Smolyak, M.S.
University of Maryland, College Park