Considerations for using predictive models that include race as an input variable: The case study of lung cancer...

Area of Science:

• Biomedical Informatics
• Health Equity
• Predictive Modeling

Background:

• Predictive models using race can perpetuate biases and worsen health disparities.
• There's a growing movement to remove race from models, but many still include it.
• Biomedical informaticists need guidance on handling race covariates in healthcare models.

Purpose of the Study:

• To propose a pragmatic framework for using or not using race in predictive models.
• To help model users avoid inadvertently exacerbating health disparities.
• To guide the application of race in electronic health records and clinical workflows.

Main Methods:

• Case study approach using lung cancer screening.
• Development of a simple framework for race covariate consideration.
• Analysis of approaches to race inclusion/exclusion in predictive models.

Main Results:

• A framework is proposed to aid decision-making regarding race in predictive models.
• The framework addresses the complexities of using race to mitigate bias.
• Guidance is provided for integrating race considerations into clinical workflows.

Conclusions:

• A practical framework can assist users in navigating the use of race in predictive models.
• Thoughtful consideration of race is crucial for promoting health equity.
• The proposed framework supports responsible implementation of AI in healthcare.