Professor Elizabeth Williamson
Professor of Biostats and Health Data Science
I studied mathematics at King's College Cambridge, followed by a Masters degree in Medical Statistics at the University of Leicester. I then moved to London to begin a PhD at LSHTM under the supervision of James Carpenter, which was awarded in 2007. For the next 7 years I worked in Australia, beginning at the NHMRC Clinical Trials Unit at Sydney University, then moving to the Murdoch Childrens Research Institute in Melbourne, followed by positions at Monash University and the University of Melbourne. In 2014, I returned to LSHTM.
My work focuses on the use of electronic health records data for health research. Initially, this work focused on missing data within causal analyses. During the COVID-19 pandemic, I was involved in the OpenSAFELY collaborative, a new platform for linked primary care record data, and contributed to using these data to answer a number of urgent health related questions through the pandemic. Since then, my methodological interests have broadened to encompass statistical challenges of both causal analyses and risk prediction using data from electronic health records.
Much of my current work explores methods related to the high dimensional propensity score, to address confounding in electronic health record data, and prediction under intervention in electronic health record data. I also have an interest in epidemiological study design and the interface between efficient study design and optimal statistical analysis.
Affiliations
Centres
Teaching
I am co-organising the module Statistics for Health Data Science for the MSc Health Data Science, as well as contributing to teaching on additional modules within the same MSc: Analysis of Electronic Health Record data and Machine Learning.
I have successfully supervised eight doctoral students to completion, and currently supervise five doctoral students.
Research
I am particularly interested in the methods used to make causal statements and to generate clinical risk predictions using routinely collected data. These sorts of data bring many challenges, including measured and unmeasured confounding, time-varying exposures and confounding (for causal analyses), changes in medication use over time and measurement error. A huge range of statistical techniques have been proposed to deal with these challenges, including extended implementations of multiple imputation, propensity scores, marginal structural models, series of sequential trials, for the causal analyses, and causal risk prediction approaches for risk prediction. Understanding whether these techniques provide valid results in real life scenarios, and identifying situations in which the simpler approaches are sufficient is critical to allow researchers to draw valid and robust conclusions from these data.
Key collaborators at LSHTM include: James Carpenter, Ian Douglas, Clemence Leyrat, John Tazare, Andriana Kostouraki, Jenni Williamson and the Beyond Cancer group led by Krishnan Bhaskaran.
Selected Publications
