Publications: On Google Scholar, QM Publists or Research Gate.

Current research areas are:
   Medical Decision Support
   System Safety and Risk

Medical Decision Support

Data is widely used for decision making in medicine, but often without explicit models of knowledge. This is clearly sometime appropriate but when data is collected by observing current practice (i.e. without running an experimental trial) it is not possible to tell cause from effect other than by using knowledge; simple models often confuse cause and effect even when there is plenty of evidence to make the distinction. Bayesian networks are a good way to combine knowledge with data to build tools for decision support.

Combining Knowledge with Data. Knowledge-based models tend to include factors that cannot be measured directly, such as an underlying disease state. We have shown the advantage of including these unmeasured (or latent) variables in predictive models. We have also investigated ways to use information from meta-analysis to supplement data available for learning a predictive model.   
  • Presenting evidence for a Bayesian network. Why should a clinician trust a BN model? How can we keep a BN up to date as medical knowledge advances? We have explored these questions by creating an ontology for the evidence supporting a BN. An example shows how evidence can be viewed online. 
  • Deriving a Network from Knowledge. Knowledge elicited from experts often gives a BN structure that needs to be simplified before it can be completed. How is the link to he knowledge maintained? We have proposed a systematic technique to simplify or abstract a BN, keeping track of the knowledge at each stage.

Trauma Care

I collaborate with the Mr Nigel Tai, a vascular surgeon and member of the Trauma Science group at the Royal London Hospital to develop decision models for lower limb trauma. We have examined the early detection of coagulopathy and the viability of limbs following trauma. this work has been carried out by Barbaros Yet, a PhD student in EECS and Zane Perkins, a trainee surgeon and PhD student.
Selected Papers

Yet, Barbaros; Perkins, Zane; Fenton, Norman; Tai, Nigel; Marsh, William; Not just data: A method for improving prediction with knowledge, Journal of biomedical informatics, 2013.  DOI: 10.1016/j.jbi.2013.10.012
Yet B, Perkins ZB, Rasmussen TE, Tai NR, and Marsh DWR (2014). Combining Data and Meta-analysis to Build Bayesian Networks for Clinical Decision Support Journal of Biomedical Informatics. DOI: 10.1016/j.jbi.2014.07.018

Forensic Psychiatry

In work led by Professor Jeremy Coid, we are using Bayesian networks for risk assessment and management in forensic psychiatry in a collaboration with the Forensic Violence Prevention research unit (VPRU) in the Wolfson Institute of Preventive Medicine, Barts and The London School of Medicine and Dentistry.  Using data collected by the VPRU, we have shown both that knowledge-based BN model both exceed the predictive accuracy of conventional models and are more useful in practice.