Ensuring Fairness in Machine Learning to Advance Health Equity

Area of Science:

• Healthcare technology
• Artificial intelligence in medicine
• Health equity research

Background:

• Machine learning (ML) models are increasingly used in clinical settings for diagnosis, treatment selection, and operational efficiency.
• Historically collected data for ML training may contain human and structural biases, posing risks to protected groups through incorrect predictions or resource allocation.
• Existing ML applications risk exacerbating healthcare disparities due to biased data and model interactions with clinicians and patients.

Purpose of the Study:

• To analyze how ML model design, data biases, and implementation can worsen healthcare disparities.
• To advocate for the proactive use of ML systems to advance health equity.
• To guide the incorporation of distributive justice principles into ML model development, deployment, and evaluation.

Main Methods:

• Review of ML model design and data bias implications in healthcare.
• Analysis of clinician-patient interactions with ML predictions.
• Description of technical implementations of distributive justice principles (equality in outcomes, performance, resource allocation).

Main Results:

• ML models trained on biased historical data can perpetuate and amplify existing healthcare disparities.
• Passive guarding against ML harms is insufficient; proactive strategies are needed to ensure equitable benefits.
• Distributive justice principles offer a framework for designing and evaluating ML systems to promote fairness.

Conclusions:

• Machine learning systems must actively incorporate principles of distributive justice to ensure equitable benefits for all patients.
• Technical implementations focusing on equality in patient outcomes, performance, and resource allocation are essential.
• Clinicians require guidance on prioritizing different justice principles based on specific contexts to mitigate ML-induced disparities.