Developing and Applying the BE-FAIR Equity Framework to a Population Health Predictive Model: A Retrospective...

Area of Science:

• Health Informatics
• Health Equity
• Predictive Modeling

Background:

• Population health programs utilize predictive models for resource allocation.
• Existing models risk perpetuating biases, worsening health disparities for marginalized communities.

Purpose of the Study:

• To develop and test the Bias-reduction and Equity Framework for Assessing, Implementing, and Redesigning (BE-FAIR) in healthcare predictive models.
• To minimize bias and enhance equity in a population health predictive model within a large health system.

Main Methods:

• A retrospective cohort study was conducted at an academic medical center from September 2020 to October 2022.
• Electronic health records and claims data were used to identify patients, stratified by race-ethnicity, gender, and social vulnerability (Healthy Places Index).
• The BE-FAIR framework was implemented, incorporating an anti-racism lens, de-siloed teams, historical intervention review, disaggregated data analysis, and calibration evaluation.

Main Results:

• Calibration of the predictive model varied significantly across demographic groups.
• Predicted hospitalization probabilities were notably lower for African Americans, Hispanics, AAPI, and multi-race individuals compared to white/Caucasians.
• Lower Healthy Places Index scores correlated with significantly lower predicted hospitalization probabilities, indicating bias.

Conclusions:

• The BE-FAIR framework provides a practical methodology for identifying and mitigating bias in healthcare predictive models.
• Implementation of BE-FAIR can improve the fairness and equity of healthcare delivery, especially for minoritized populations.
• This framework supports the development of more inclusive and effective population health strategies.