Considerations for addressing bias in artificial intelligence for health equity

Area of Science:

• Healthcare technology
• Biomedical informatics
• Health equity

Background:

• Health equity is a key goal for diverse healthcare stakeholders.
• Digital health technologies, particularly AI/ML, offer potential to improve equitable access to diagnosis and treatment.
• However, AI/ML may also exacerbate existing health disparities if bias is not carefully managed.

Purpose of the Study:

• To propose an expanded Total Product Lifecycle (TPLC) framework for healthcare AI/ML.
• To describe sources and impacts of undesirable bias in AI/ML systems across all lifecycle phases.
• To educate stakeholders on identifying and mitigating AI/ML bias to ensure health equity.

Main Methods:

• The study proposes an expanded TPLC framework for healthcare AI/ML.
• It outlines methods for analyzing undesirable bias using appropriate metrics.
• It discusses potential strategies for mitigating identified biases.

Main Results:

• The proposed framework details bias sources and impacts within each AI/ML TPLC phase.
• It emphasizes the need for metrics to analyze bias.
• It suggests mitigation strategies to address inequities.

Conclusions:

• Addressing bias in AI/ML is crucial for achieving health equity.
• The expanded TPLC framework provides a roadmap for stakeholders to manage AI/ML bias.
• Proactive identification and mitigation of bias will lead to better health outcomes for all populations.