Statistician interested in how prognosis research is conducted
My research interests are focused on methodological aspects of developing and validating multivariable prediction models. Risk prediction models can predict how a person’s disease is likely to develop (their prognosis) or their probability of developing a disease in the future, using their current characteristics like weight and blood pressure. I use applied statistics, systematic review and simulation techniques to investigate how prediction models are usually developed and validated, and how we can improve this methodology for more useful, accurate models.
Prediction models are developed using information from a group of patients, selecting the important characteristics and working out how they are associated with the outcomes of interest. After the model has been developed, it is validated using information from another, similar group of patients, to check that it works for all relevant patients. I am using an independent external validation dataset to investigate how a risk model’s performance is affected by missing data in, and the sample size of, its development dataset. I am also looking at how missing data should be dealt with when developing and validating a model.
I am also interested in how well prediction model studies are reported in the scientific literature. When important details are left out of a research article, we cannot judge how well the study was carried out and how useful the resulting model is. The TRIPOD reporting guideline was developed to help researchers fully report their prediction modelling studies. We are investigating how well prediction model studies in diabetes follow TRIPOD, to give us a baseline assessment of the quality of reporting in this area.
Most recently, I have been looking at machine learning approaches and compare them to more traditional approaches for prediction.
I graduated with a BSc in Mathematics from the Royal Holloway University of London in 2013, followed by an MSc in Statistics (Medical) from UCL in 2014. My MSc thesis is entitled “How should we choose the shrinkage parameter when using penalized regression to develop risk models”. I joined the Centre for Statistics in Medicine in 2015.
Wynants L. et al, (2020), Bmj, m1328 - m1328
Christodoulou E. et al, (2019), J clin epidemiol, 110, 12 - 22
Collins GS. et al, (2016), Current anesthesiology reports
Assessing the impact of handling missing data when validating a prognostic model.
Ma J. et al
Clinical prediction models: a critical review of online risk calculators
Ma J. et al