Adnan and Erik discussed their ongoing projects that aim to transform healthcare delivery through artificial intelligence (AI). Adnan’s project focuses on automating the screening process for diabetic retinopathy — a condition caused by diabetes that can lead to vision loss. Currently, millions of retinal images are analyzed manually by trained technicians each year in the UK, which is time-consuming and costly. His team, in collaboration with City St George’s, University of London and Moorfields Eye Hospital, is validating AI algorithms to automate parts of this process. Once validated, these algorithms will enable faster, more cost-effective diagnosis and potentially lead to one of the first large-scale AI deployments within the NHS.
Erik’s work uses AI to analyze patient feedback from the NHS Friends and Family Test, which generates thousands of monthly comments that would otherwise go unread. Through natural language processing, key themes are identified and shared with quality improvement teams. This approach has since been scaled across multiple NHS organizations, with the code made freely available. Additionally, Erik’s team has developed a secure data environment, iCARE, capable of processing billions of structured and unstructured health records in real time. This has created an AI testbed for projects such as developing automated, setting-specific discharge summaries, advancing the use of generative AI in healthcare.
Both experts highlighted key lessons in scaling AI within the NHS. Adnan emphasized that successful implementation depends on strong digital infrastructure and close collaboration with IT departments to address regulatory concerns like GDPR. Erik pointed out the importance of executive-level leadership and local ownership of AI tools. Projects succeed when organizations adapt AI solutions to their unique needs rather than simply adopting a one-size-fits-all model.
On industry partnerships, both stressed the need for collaboration based on solving clearly defined healthcare problems rather than retrofitting existing tools. Adnan noted that a well-defined problem, such as ensuring AI performs as safely and effectively as human screeners, encourages healthy competition and innovation among providers. Erik added that NHS, academia, and industry should co-design solutions, with the NHS taking a stronger role in articulating specific needs to foster productive partnerships.
Regarding development and monitoring, Erik underscored the need for secure data environments where healthcare, academia, and industry can collaborate to build evidence for regulatory approval. Adnan cautioned against inconsistencies in vendors’ studies, suggesting a framework similar to drug trials is needed to ensure safety and reliability. Both discussed the challenge of regulating AI systems that continuously learn and evolve. They proposed running a separate, parallel learning model to ensure patient safety while allowing innovation to continue.
They also reflected on broader challenges, particularly the impact of AI on human behavior and workflows. Erik highlighted concerns around tools like digital scribes that record entire patient interactions, which could alter communication dynamics and require additional clinician oversight. Adnan reiterated the importance of focusing on clear problem statements and simple, safe technology applications. Together, they concluded that while AI continues to evolve rapidly, its implementation in healthcare must remain gradual, collaborative, and centered on patient outcomes — guided by the principle of “think big, start small.”
