Rory Ormiston

Fellowship Bio

Trauma & Orthopaedic Registrar and PhD researcher at the University of Southampton, developing clinically useful AI to improve shared decision-making. My interests include machine learning, medical imaging, perioperative care and translating AI safely into NHS practice.

Fellowship Project

Every patient diagnosed with oesophageal or gastric cancer is discussed by a multidisciplinary team (MDT) of surgeons, oncologists, radiologists, pathologists and specialist nurses before treatment is recommended. The aim of this project was to implement the use of a validated machine learning algorithm which predicts MDT treatment decisions. Rather than attempting to replace clinical decision-making, the project focused on understanding where AI could assist clinicians by streamlining workflows, identifying patterns and helping teams spend more time discussing the patients who need it most. My role was to lead the implementation of the ML algorithm. I designed the study methodology, collected detailed observational data from MDT meetings, analysed discussion times and treatment pathways, and interpreted the findings. The project has successfully characterised the current MDT workflow, quantified how meeting time is distributed across different patient groups and treatment decisions, and identified several challenges to implementation;

• Predictive accuracy does not equate to implementation success,
• Clinical practice evolves and algorithms must too,
• MDT decisions are iterative but are being modelled as a point-in-time decision,
• Missing and ambiguous data challenges AI more than clinicians.

These findings have been presented at scientific meetings and are being prepared for publication. The next phase will use these data to develop and evaluate AI-assisted MDT tools, assess their impact on workflow efficiency and decision-making, and explore how they can be safely integrated into routine clinical practice while maintaining clinician oversight and patient-centred care.

Fellowship Testimonial

This fellowship has refined how I think about artificial intelligence in healthcare. I leave with a much greater appreciation that successful implementation is often a bigger challenge than model development itself and that the NHS is arguably the world’s largest brownfield digital transformation project. Reading Digital Transformation at Scale emphasised to me that rather than trying to design the perfect solution from the outset, small-scale implementation exposes the practical challenges that only become apparent in real clinical environments. I experienced this while attempting to implement our OC MDT model. The model was trained on retrospective, complete data, limiting its ability to generalise to routine NHS practice. That experience has directly influenced how I will design future models during my PhD, with a greater emphasis on available and representative datasets from the outset. Implementation highlighted further challenges. Registry datasets capture outcomes but rarely the reasoning behind decisions. MDT discussions are iterative rather than point-in-time classifications, and missing or ambiguous data challenges AI far more than clinicians. Successful clinical decision support therefore depends on thoughtful workflow integration, not simply predictive accuracy. One analogy that has stayed with me came from Dr Keith Grimes, who described AI as the “Wild West”: full of opportunity, but requiring responsible people with strong values to guide its safe adoption until regulation catches up. Finally, I remain convinced that clinicians developing AI should understand the code behind the models they use. Continuing to improve my programming skills will therefore remain a priority as I enter the final year of my PhD, supported by an ORUK/RCSed Research Fellowship.