To evaluate the impact of removing race and ethnicity from the Breast Cancer Surveillance Consortium advanced breast cancer risk model on algorithmic bias and model equity, specifically addressing how this removal affects prediction accuracy across different racial and ethnic groups.
Key Findings:
Excluding race and ethnicity led to overestimation of risk in Asian women (expected/observed = 1.28 vs. 1.00) and underestimation in Black women (expected/observed = 0.61 vs. 1.00).
Advanced breast cancer risk among Asian women classified as intermediate/high risk increased from 6.1% to 16.7%.
The percentage of Black women identified as intermediate/high risk decreased from 75.3% to 47.5%.
The model excluding race and ethnicity showed worse calibration, potentially leading to suboptimal screening strategies.
Interpretation:
Removing race and ethnicity from the risk model negatively impacts the accuracy of risk predictions, particularly for Asian and Black women, which could exacerbate existing health disparities in breast cancer outcomes.
Limitations:
The study may not account for all factors influencing breast cancer risk, such as socioeconomic status, genetic predispositions, and environmental factors.
The findings are based on a specific cohort and may not be generalizable to all populations, particularly those outside the study demographics.
Conclusion:
The inclusion of race and ethnicity in breast cancer risk prediction models is crucial for accurate risk assessment and effective screening strategies, highlighting the need for tailored approaches in clinical practice.
