Year 3 Progress Update

During the reporting period, AI‑POD made strong advances in bringing AI‑driven innovation closer to real‑world clinical practice. Two major milestones were achieved in the development of the project’s multimodal cardiovascular risk prediction framework. The first was the delivery of the initial AI‑POD cardiovascular risk score, integrating photon‑counting CT imaging with clinical, laboratory, and lifestyle data. Advanced AI methods enhance predictive power while maintaining transparency, positioning the score firmly as a decision-support tool that strengthens clinical judgment. Building on this achievement, the consortium also advanced the development of a second-generation AI‑POD score algorithm, incorporating novel AI approaches for the integration of imaging, clinical, laboratory, and report-derived information. Together, these key developments represent important steps towards more accurate, robust, and explainable cardiovascular risk prediction.

Complementing these efforts, AI‑POD delivered the AI‑POD Feedback Cockpit, which enables the system to learn directly from clinical use. By tracking how clinicians interact with AI recommendations, collecting structured and free‑text feedback, and monitoring usage of the CDSS and Citizen App, AI‑POD now operates with a closed feedback loop that continuously feeds real‑world insights back into system development.

At the same time, a major achievement was the completion of a series of interviews, surveys and engagement activities to systematically capture stakeholder perspectives on the use of AI in cardiovascular risk prediction. Both strong support for multimodal, personalised AI tools and clear expectations regarding explainability, fairness, and integration into existing care pathways were identified. Concerns around bias, digital exclusion, and risk communication were explicitly addressed, ensuring that ethical and societal considerations actively informed technical and clinical decisions rather than remaining theoretical.

In addition, findings from the AI‑POD stakeholder webinar and the accompanying post‑webinar survey demonstrated a high level of interest in AI‑driven cardiovascular risk prediction across diverse stakeholder groups, with particularly strong engagement from clinicians and clinical managers. While participants expressed enthusiasm about the clinical potential of AI‑POD tools, they consistently emphasised the need for practical usability, seamless integration into existing workflows, and clearly demonstrated clinical value. Feedback highlighted the importance of realistic implementation pathways, transparent governance structures, and AI systems that complement and support routine clinical practice rather than adding complexity or disruption.

Moreover, in response to clinical, technical, and regulatory complexities and constraints, the consortium repositioned the prospective clinical work as a feasibility study comprising 300 patients. This adapted study design places a strong emphasis on assessing workflow integration in routine clinical environments, ensuring robust multimodal data acquisition and harmonisation, and evaluating system usability and real‑world interaction with AI‑based tools for both clinicians and patients. By October 2026, 150 participants (50%) had been successfully recruited across participating clinical sites, with targeted mitigation measures implemented to support continued enrolment.