Eliminating Algorithmic Racial Bias in Clinical Decision Support Algorithms: Use Cases from the Veterans Health...

Area of Science:

• Health Equity
• Clinical Informatics
• Health Services Research

Background:

• Clinical algorithms and predictive models can perpetuate health disparities.
• The Veterans Health Administration (VHA) is committed to identifying and mitigating bias in its healthcare tools.
• Previous use of race modifiers in clinical equations has raised concerns about equity.

Purpose of the Study:

• To examine and eliminate the use of potentially biased clinical equations and predictive models within the VHA.
• To detail the processes, successes, challenges, and future steps in addressing algorithmic bias.
• To advance racial health equity in clinical decision support algorithms.

Main Methods:

• Applying equity- and evidence-based principles to review clinical tools.
• Eliminating the race modifier for estimated kidney function calculations.
• Implementing equity lenses in the Stratification Tool for Opioid Risk Mitigation (STORM) and Care Assessment Need (CAN) predictive models.
• Developing strategies for more equitable pulmonary function testing measurements.

Main Results:

• Successful elimination of the race modifier in estimated kidney function, a significant step toward reducing racial bias.
• Progress in applying equity considerations to pulmonary function testing.
• Demonstrated application of equity lenses in STORM and CAN predictive models, improving their fairness.
• Identification of key processes and challenges in algorithmic bias mitigation.

Conclusions:

• The VHA's systematic approach demonstrates a commitment to advancing racial health equity.
• Eliminating biased components in clinical algorithms is crucial for equitable healthcare delivery.
• Continued vigilance and adaptation of clinical decision support tools are necessary to ensure fairness and improve patient outcomes.