Co-design for Trustworthy AI: An Interpretable and Explainable Tool for Type 2 Diabetes Prediction Using Genomic Polygenic Risk Scores

Explainability addresses a critical gap in genetic risk assessment: The paper identifies that polygenic risk scores (PRS), while effective at identifying individuals at high risk for type 2 diabetes, have lacked interpretability tools. XPRS addresses this by breaking down risk scores into specific gene-level and SNP (single-nucleotide polymorphism) contributions, allowing clinicians and patients to understand which genetic factors are driving an individual's risk profile.

Co-design with multiple stakeholders is essential for trustworthy AI in healthcare: Rather than developing the tool in isolation, the researchers used a structured co-design approach incorporating legal, medical, ethical, and technical perspectives from the outset. This multi-disciplinary approach helps ensure the tool serves actual user needs and addresses real-world concerns before deployment in clinical settings.

Formal ethical frameworks can guide responsible AI development: The study piloted established governance tools including Z-inspection (an ethically aligned co-design methodology) and the Council of Europe's HUDERIA framework. This demonstrates that responsible AI isn't just about technical performance—it requires systematic evaluation against human rights, democracy, and rule of law standards.

Transparency in genetic risk tools has implications for patient autonomy and equity: By making genetic contributions visible and understandable, XPRS enables more informed decision-making by patients and clinicians. However, this also raises important questions about how genetic information is communicated, who has access to it, and whether such tools might inadvertently reinforce health disparities if not carefully implemented across diverse populations.