Participant flow diagrams for health equity in AI

Area of Science:

• Medical research methodology
• Artificial intelligence in healthcare
• Health equity

Background:

• Selection bias is a significant issue in medical research, often stemming from underrepresentation of disadvantaged populations.
• The use of non-representative samples in artificial intelligence (AI) and machine learning (ML) can exacerbate these biases in clinical algorithms.

Purpose of the Study:

• To propose the addition of participant flow diagrams in AI studies for enhanced transparency.
• To detail sociodemographic and clinical characteristics of excluded participants across study phases.
• To identify and mitigate potential algorithmic biases before clinical implementation.

Main Methods:

• Advocating for a standardized participant flow diagram model for AI studies.
• Building upon the 'Data Cards' initiative for AI research transparency.
• Presenting a case study for practical implementation of the proposed flow diagram.

Main Results:

• The proposed participant flow diagram aims to systematically document participant exclusion.
• Standardized reporting can reveal inequities embedded within AI applications.
• Identifying potential biases early can lead to more reliable algorithms.

Conclusions:

• Participant flow diagrams are crucial for identifying selection bias in AI research.
• Implementing these diagrams can facilitate the development of more equitable and reliable clinical algorithms.
• This approach promotes transparency and fairness in AI-driven healthcare.