Easy ensemble classifier-group and intersectional fairness and threshold (EEC-GIFT): a fairness-aware machine...

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Public Health

Background:

• Lung cancer screening (LCS) eligibility criteria can exhibit racial bias.
• Real-world data is crucial for developing equitable healthcare tools.

Purpose of the Study:

• To develop an accurate and racially unbiased LCS eligibility mechanism.
• To address disparities in lung cancer screening access.

Main Methods:

• A fairness-aware machine learning framework was created using the Group and Intersectional Fairness and Threshold (GIFT) strategy.
• The framework integrated with Easy Ensemble Classifier (EEC) or Logistic Regression (LR) models.
• Performance was compared against the 2021 USPSTF criteria and PLCOM2012 model using the Prostate, Lung, Colorectal, and Ovarian (PLCO) cancer screening trial data.

Main Results:

• The EEC-GIFT* and LR-GIFT* models demonstrated improved sensitivity without sacrificing specificity compared to the 2021 USPSTF criteria.
• These models showed comparable predictive performance (Area Under the Curve) to the PLCOM2012 model.
• The EEC-GIFT* model effectively eliminated racial bias (equal opportunity difference between Black and White smokers), while other models showed significant bias.

Conclusions:

• The EEC-GIFT* LCS eligibility mechanism significantly mitigates racial bias in determining screening eligibility.
• This approach maintains high predictive accuracy, offering a more equitable LCS tool.