A scoping review of fair machine learning techniques when using real-world data.

Area of Science:

• Health Informatics
• Artificial Intelligence in Medicine
• Machine Learning for Healthcare

Background:

• Artificial intelligence (AI) and machine learning (ML) are increasingly used in healthcare for clinical decision support.
• Concerns exist regarding fairness and bias in AI/ML, potentially exacerbating health inequities due to uneven distribution of benefits and drawbacks.
• Algorithmic bias can disproportionately impact certain subpopulations, widening existing health disparities.

Purpose of the Study:

• To summarize existing literature on tackling algorithmic bias and optimizing fairness in AI/ML models.
• To identify research gaps concerning the use of real-world data (RWD) in healthcare AI/ML fairness.
• To provide insights for researchers in the burgeoning field of fair AI/ML in healthcare.

Main Methods:

• Conducted a thorough review of techniques for assessing and optimizing AI/ML model fairness in healthcare using RWD.
• Focused on appraising quantification metrics for fairness assessment, publicly accessible datasets for ML fairness research, and bias mitigation approaches.

Main Results:

• Identified 11 papers focused on optimizing model fairness in healthcare applications.
• Found limited current research on mitigating bias in RWD, both in disease variety and healthcare applications.
• Noted that pre-processing techniques often show positive outcomes, but research gaps remain in understanding bias root causes and multi-modal data.

Conclusions:

• Fair AI/ML in healthcare is a developing field requiring increased research focus.
• Further research is needed to address bias root causes, broaden fairness research with RWD, and evaluate bias in multi-modal data.
• This review offers valuable insights and identifies critical gaps for researchers in AI/ML fairness using real-world healthcare data.