It was great to welcome back DRIVE-Health PhD student, Dr Hugh Logan-Ellis - a Diabetes and Endocrinology Registrar at King's and ex-Research Fellow in the Department of Medicine at Dalhousie University - who delivered our September Seminar Series. In his talk “Extracting Clinical Value from EHR Data: Challenges, Pitfalls, and Practical Lessons" , Hugh shared what clinicians have taught him about the reality of working with Electronic Health Record data and what they genuinely need from #AI tools, rather than what researchers might think they should want. Hugh has learned that making the most clinically useful tool could matter more than theoretical perfection. He'll discuss some principles he's gathered to help create AI solutions that fit seamlessly into clinical workflows, which he hopes might help others bridge the gap between academic research and genuine patient benefit. Using his PhD research on creating a single unit of health from #EHR data as a central example, Hugh will explore broader challenges: the messiness of real-world clinical data, the proliferation of unused risk scores, and why so many promising algorithms never make it past publication. These insights aim to help researchers develop tools that won't just die in papers, but have a real chance of improving clinical care. Seminar Series Event: " Extracting Clinical Value from EHR Data: Challenges, Pitfalls, and Practical Lessons" Date and Time: Thursday 25 September 2025, 12:00 – 13.00 hrs (BST) Location: The Judy Dunn Room, SGDP Building, Denmark Hill Campus, London, SE5 8AF Attendance: Mandatory for all DRIVE-Health students, therefore please accept the calendar invitation. Registration: Alumni and wider King's College London research community all welcome - please email drive-health-cdt@kcl.ac.uk to let us know if you would like to attend. Abstract: Picture the scene: It's Saturday morning, you're the senior resident doctor on call in a busy hospital, and you have a 40-page list of patients due for review. Half of your junior colleagues have called in sick, and you know you can't possibly see everyone. How do you decide who needs to be seen most urgently? The information to make these decisions is in the electronic health records, but accessing it quickly means opening each patient's chart individually. My PhD tries to tackle this problem: could we use an algorithm to compress scattered clinical data into a single, practical number? This question has led me on an interesting journey. I've spoken with clinicians from around the world about how they decide who is "sickest," discovering a surprising variety of terms for essentially the same idea and realising we might need more than one measure. My research has taken me to Canada to collaborate with Professor Kenneth Rockwood OC, whose groundbreaking work on frailty measurement has significantly shaped clinical practice worldwide. Working alongside him has given me valuable insights into why some academic ideas successfully transform patient care, while others remain confined to journals. As I explored increasingly sophisticated approaches to measure sickness, from simple laboratory-based indices to complex machine learning models, I stumbled across a key insight. Supervised machine learning can hindered by retrospective health data because when sick patients are successfully treated, they don’t have poor outcomes. This isn't just a quirky finding relevant to my PhD; it has broader implications for using a supervised paradigm on retrospective data whenever effective treatments are already in place. Bio Hugh is a resident medical doctor specialising in Internal Medicine and Diabetes and Endocrinology, working on his PhD at King's College London. His research focuses on measuring patient health status using electronic health records, drawing on his experience working across various healthcare settings in the UK and internationally.