William Marsh is a Senior Lecturer in Computer Science at Queen Mary University of London and a member of the Risk and Information research group. He has played a major role in the group's previous work on decision support for medical applications, working with Bayesian networks (BNs) and causal modelling. He has also pioneered the use of BNs in railway risk analysis. He enjoys collaborating with clinicians and has worked on applications in trauma care, forensic psychiatry, musculoskeletal injury, diabetes and rheumatology. His research interests are in better ways to build Bayesian networks suitable for decision support, effective ways to combine knowledge and data and ways to improve the usefulness of predictions to decision makers (e.g. explanation, explicit knowledge).
Before joining Queen Mary he worked in software development and assessment and led software and safety assessment projects in avionics and railway engineering. He has a PhD from Southampton University.
Dr Marsh collaborates with clinical professionals to develop, test and deploy practical decision-making tools. His current projects focus on chronic medical conditions and the potential for decision support to enable greater patient autonomy (PamBayesian), improved triage in musculoskeletal injury (RealmAI) and trauma care (Trauma Models).
Prediction in medicine (diagnosis, prognosis, the effect of treatment) is expected by many to transform health care. However, so far the practical impact has been limited (except perhaps in relation to diagnostic imaging). There are many practical challenges to overcome to change this, with questions such as: How do we know which predictions are likely to have the biggest impact? How do we capture and model medical knowledge efficiently? How can the very complex clinical data records in the Electronic Health Record (EHR) systems now used by hospitals and GPs be used routinely so that a prediction system improves with experience (like a real doctor)?