The selective deployment of AI in healthcare: An ethical algorithm for algorithms

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Health Equity

Background:

• Machine-learning algorithms offer significant potential in healthcare diagnostics and prognostics.
• Performance disparities exist for underrepresented groups due to data deficits, potentially causing harm.
• Delaying deployment for perfect equity risks avoidable deaths in well-represented populations.

Purpose of the Study:

• To address the ethical dilemma between algorithmic equity and utility in healthcare.
• To propose a framework for the selective deployment of diagnostic and prognostic tools.
• To examine case studies in breast cancer and melanoma to inform ethical deployment strategies.

Main Methods:

• Analysis of two case studies: breast cancer and melanoma.
• Ethical argumentation on selective deployment versus universal delay.
• Proposal of non-algorithmic measures to address historic injustice and data deficits.
• Development of an ethical algorithm for deployment decisions.

Main Results:

• Selective deployment for well-represented groups can be ethically justifiable when inclusion causes harm to underrepresented groups.
• Non-algorithmic measures like transparency, additional services, and data collection are crucial for excluded populations.
• Urgent commitments and regulation are needed to minimize delays for excluded groups.

Conclusions:

• A balanced approach is necessary, allowing selective deployment while actively working towards universal, equitable algorithmic accuracy.
• Transparency with patients regarding algorithmic limitations is paramount.
• Addressing historic injustice requires dedicated efforts beyond algorithmic adjustments, supported by policy and funding.