Katharine Wykes

Fellowship Bio

Dual Anaesthetic & Intensive Care Registrar, Oxford University Hospitals NHS Trust. Previously a Technical Director in software engineering, specialising in system architecture, build and deployment, and back-end database integration. Proficient in 8 programming languages, including Python.

Fellowship Project

Doctors in A&E and Intensive Care make split-second calls, piecing together accurate information quickly, often with lives at stake. Part of that comes from reviewing a patient’s journey through hospital, so lessons can be learned; known as a morbidity and mortality (M&M) review. AI promises to help, but many tools are expensive commercial products, or have never been rigorously tested in practice. My fellowship involved evaluating eight open-source models for detecting pneumothoraces and fractures on real NHS X-rays, and testing whether LoRA (an efficient training technique) can improve accuracy cheaply. I also built an M&M tool (AutoMM) that summarises a patient’s clinical and physiological journey through Intensive Care.

My contributions: Attended a weekly MDT throughout. Wrote up the open-source model publication, now revised for resubmission. Wrote data agreements for the commercial comparison study, with vendor relationships to resume in Q4. Wrote and tested a Python class for LoRA adaptation, and networked with Big Data Institute engineers. Submitted the necessary governance documentation for AutoMM, documented the previously undocumented ICU database schema, and built the Python query classes (AutoMM study accepted at ESICM LIVES Lisbon). Authored server/network specs for OxCAIR’s hardware. Secured an NIHR PreDoctoral Fellowship.

Milestones and next steps: Achieved: open-source paper submitted and revised; LoRA testing progressed; ICU schema documented; DPIA submitted; ESICM poster accepted; server specs written; PreDoctoral Fellowship secured. Next steps: resubmit publication; complete LoRA evaluation; resume commercial study; gain DPIA approval; build AutoMM presentation layer; deploy server hardware; begin PreDoctoral study.

Fellowship Testimonial

This fellowship gave me more than I expected. It gave me the opportunity to build genuine professional relationships with senior people across the organisation, from clinical governance leads to the OxCAIR team, connections that have already shaped my career beyond this year. I enjoyed using my programming skills within a real, complex NHS environment, with all the challenges that brings. Utilising my previous experience to interrogate the Intensive Care database and formally document the data schema, which was useful for the AutoMM work, and will be useful for future data-driven projects. Equally rewarding was moving beyond researching LoRA as a concept to actually coding an adaptation layer to fine-tune the open source models, and beginning to evaluate whether the adaptation genuinely improved performance. I’m proud to be graduating having secured an NIHR PreDoctoral Fellowship, so my digital AI work will continue, with a research team active in Oxford, pushing the boundaries of machine learning within infectious diseases and genomic sequencing. I will be focusing on antimicrobial stewardship within ICU using statistical, machine-learning, and deep-learning approaches, a natural continuation of the skills and relationships this fellowship has built. Looking back, this year has changed how I see my own career, not as a choice between clinical practice and research, but as a chance to bring both together. I’m leaving with data governance skills/knowledge that I didn’t have before, a research network I’ll keep building on, and belief that I can help bring robust and rigorously evaluated AI into the NHS to try to genuinely improve care.