Can large language models be privacy preserving and fair medical coders?

TL;DR

This paper investigates the challenges of applying differential privacy to medical coding with large language models, highlighting significant impacts on model performance and fairness between genders.

Contribution

It provides an empirical analysis of privacy-utility and privacy-fairness trade-offs in privacy-preserving NLP models for healthcare.

Findings

Over 40% reduction in micro F1 scores with DP

Over 3% increase in gender recall gap due to DP

Highlights challenges in deploying privacy-preserving models in healthcare

Abstract

Protecting patient data privacy is a critical concern when deploying machine learning algorithms in healthcare. Differential privacy (DP) is a common method for preserving privacy in such settings and, in this work, we examine two key trade-offs in applying DP to the NLP task of medical coding (ICD classification). Regarding the privacy-utility trade-off, we observe a significant performance drop in the privacy preserving models, with more than a 40% reduction in micro F1 scores on the top 50 labels in the MIMIC-III dataset. From the perspective of the privacy-fairness trade-off, we also observe an increase of over 3% in the recall gap between male and female patients in the DP models. Further understanding these trade-offs will help towards the challenges of real-world deployment.