Addressing algorithmic bias in lung cancer screening eligibility.

Area of Science:

• Public Health
• Health Equity
• Medical Informatics

Background:

• Lung cancer screening guidelines, including USPSTF, are based on predominantly White populations.
• Existing strategies exhibit algorithmic bias, performing inconsistently across diverse racial groups.
• This inconsistency suggests a need to re-evaluate eligibility criteria for equitable application.

Purpose of the Study:

• To assess the performance of current lung cancer screening eligibility strategies in a predominantly Black/African American cohort.
• To explore methods for resolving algorithmic bias to enhance equity in lung cancer screening eligibility.
• To propose and evaluate novel race-aware strategies.

Main Methods:

• Utilized the Southern Community Cohort Study (predominantly Black/African American participants).
• Evaluated eight existing lung cancer screening eligibility strategies and two new race-aware strategies.
• Compared performance metrics such as sensitivity and false positive rates across racial groups.

Main Results:

• Existing strategies identified fewer Black/AA participants as eligible for lung cancer screening compared to White participants.
• Removing race or other criteria did not resolve racial disparities in eligibility.
• Race-aware strategies demonstrated the best sensitivity-specificity trade-off, minimizing inequities.

Conclusions:

• Algorithmic bias in lung cancer screening eligibility is evident in current guidelines.
• Race-aware approaches are crucial for addressing disparities and ensuring equitable access to lung cancer screening.
• Tailoring eligibility thresholds by race can improve fairness and effectiveness.