Evaluation and Mitigation of Racial Bias in Clinical Machine Learning Models: Scoping Review.

Area of Science:

• Medical Artificial Intelligence
• Algorithmic Fairness
• Health Disparities

Background:

• Racial bias in clinical machine learning (ML) models is a significant concern with the potential to exacerbate health disparities.
• Current understanding and best practices for addressing racial bias in clinical ML are not well-established.

Purpose of the Study:

• To conduct a scoping review to characterize methods for assessing racial bias in ML.
• To identify strategies for enhancing algorithmic fairness in clinical ML applications.

Main Methods:

• A comprehensive scoping review was performed following PRISMA Extension guidelines.
• Literature search across PubMed, Scopus, Embase, and Google Scholar identified 12 relevant studies from 635 records.

Main Results:

• ML applications varied across diagnosis, outcome, and clinical score prediction using diverse data types.
• Racial bias was reported in 67% of studies; fairness metrics were inconsistent.
• Bias mitigation strategies, primarily preprocessing methods, successfully improved fairness in all implementing studies.

Conclusions:

• Increased emphasis on evaluating and mitigating racial bias in clinical ML is crucial due to its widespread use and potential patient harm.
• Inconsistent adoption of algorithmic fairness principles is hindered by poor data availability and reporting standards.
• Standardized reporting and data availability are recommended to improve transparency and facilitate bias evaluation in medical ML.