Senior Research Associate
Dr Annabel L. Davies
I am a Senior Research Associate in Evidence Synthesis at the University of Bristol. My research interests lie at the intersection of statistical physics and medical statistics. In particular I am interested in how ideas and techniques from statistical physics can contribute to the understanding and development of medical statistics methodologies. Previous work has focussed on concepts from physics such as random walks and graph theory with applications to network meta-analysis and survival analysis. In the future I hope to expand this work to include techniques such as constrained optimisation and constraint satisfaction, network science and machine learning. I welcome collaboration from physicists and statisticians and believe the two disciplines can benefit from shared ideas and expertise.
Statistical physics and network meta-analysis
In the main, my research has focused on network meta-analysis (NMA). This is a statistical technique to combine data from different clinical trials in order to make inferences about multiple competing treatments. Nodes of the network represent the treatments while linking edges represent trials comparing the treatments.
The representation of treatments and trials as a network graph has parallels with systems in physics. Examples include electrical circuits, mechanical systems, and interacting particles. Many techniques have been developed within statistical physics to analyse such networks. I am interested in exploiting these techniques to improve understanding and develop methodology in the context of network meta-analysis. For example, in previous work I used network theory to investigate how network geometry affects the performance of NMA. This led to an improved understanding of how bias can enter into the estimation procedure. In another project I, along with collaborators, developed an analogy between NMA and random walks. We used this analogy to quantify the contribution of different pieces of evidence in the network and in doing so, were able to overcome limitations of existing NMA software.
I am interested in learning about different techniques from statistical physics which have the potential to be useful for NMA. For example, those that relate to networks or systems of multiple interacting components. Topics that have piqued my interest and that I hope to investigate in future projects include constraint satisfaction & optimisation, network theory and the physics of spin systems and neural networks. I am also always looking to learn about the conceptual and practical challenges that face statisticians when utilising NMA in clinical research. Such challenges often spark interesting theoretical questions whilst also ensuring that the focus of methodological research has useful practical applications.
