Algorithmic Fairness in Clinical Natural Language Processing: Challenges and Opportunities

Abstract

The surge in research and development of clinical natural language processing (NLP) has prompted inquiries into the algorithmic fairness of the proposed and deployed technical solutions. In spite of the proliferation of research, limited work has synthesized reflected on the state of algorithmic fairness in clinical NLP. In this short paper, we summarize the findings of our scoping review of literature and present challenges and opportunities in the domain. We identify challenges and opportunities related to studying and measuring protected groups, selecting appropriate methodology, data sharing and privacy, as well as generalizability. The goal of this article is to start a discussion and raise awareness about the gaps encountered within algorithmic fairness in clinical NLP and pave the way for future research.