Addressing inaccurate racial and ethnic data in medical AI

With artificial intelligence (AI) increasingly integrated into healthcare, the inaccuracies of racial and ethnic data found in electronic health records (EHRS) can have a negative impact on patient care. As hospitals and providers struggle to collect such data consistently and accurately categorize individual patients, AI systems trained with these datasets could inherit and perpetuate racial bias.

In a new publication in PLOS Digital Health, bioethics and legal experts call for immediate standardization of how racial and ethnic data is collected, and immediate standardization for developers to ensure the quality of racial and ethnic data in healthcare AI systems. This study integrates concerns about why EHR patient race data is not accurate, identify best practices for health systems and healthcare AI researchers to improve data accuracy, and provide new templates that ensure the quality of race and ethnic data is transparent.

Author Alexandra Tsalidis, MBE, said, “If AI developers listen to recommendations to disclose how racial and ethnic data was collected, it will not only promote transparency in medical AI, but also help patients and regulators critically assess the safety of the resulting medical devices, as well as care tools.”

Racial bias in AI models is a major concern as technology is increasingly integrated into healthcare. This article provides specific methods that can be implemented to address these concerns. ”

Francis Shen, J.D., Ph.D., Senior Author

Although more work needs to be done, this article offers a starting point suggesting co-author Lakshmi Bharadwaj, MBE. “Open dialogue on best practices is a critical step, and the approach we propose can generate significant improvements.”

This study was supported by the Artificial Intelligence (Bridge2AI) program from the NIH Bridge, and the NIH Neuroethics Grant (R01MH134144).