What Is Fair? Defining Fairness in Machine Learning for Health.

Area of Science:

• Health Informatics
• Machine Learning
• Medical Ethics

Background:

• Machine learning (ML) models are increasingly used in clinical decision-making.
• Ensuring ML model safety, effectiveness, and equity is critical to prevent exacerbating health disparities.
• Understanding how ML models can lead to unfair outcomes is essential for equitable healthcare.

Purpose of the Study:

• To examine the conceptualization of fairness in ML for health.
• To investigate reasons behind unfair ML decisions in healthcare.
• To review methods for measuring fairness in real-world health applications.

Main Methods:

• Literature review of fairness notions in ML for health.
• Analysis of group, individual, and causal-based fairness frameworks.
• Discussion of operationalizing fairness in health-focused ML applications.

Main Results:

• Common fairness notions within group, individual, and causal frameworks were reviewed.
• The study identified potential sources of unfairness in ML models used in healthcare.
• Various measurement approaches for ML fairness in real-world health applications were discussed.

Conclusions:

• Fairness in ML for health requires careful conceptualization and measurement.
• Addressing fairness is key to preventing ML models from amplifying existing health disparities.
• Future research should focus on operationalizing fairness in clinical ML applications to ensure equitable patient outcomes.