Statistical physics application to evidence synthesis in medicine

Lead Research Organisation: The University of Manchester
Department Name: Physics and Astronomy

Abstract

In this project we aim to apply tools and ideas from statistical physics and the theory of complex networks to evidence synthesis methods used in medical research with particular focus on Network Meta-Analysis (NMA). This will involve mathematical modelling, statistical analysis and computational simulation of large, complex data networks with the aim of combining results from medical trials in a way that improves the precision and statistical power of treatment effect estimates. Our first avenue of investigation will be developing methodology for planning future research based on the results of a NMA of existing trials. This has vital application in publicly and privately funded medical research as it will inform clinicians on the most cost effective and efficient way to gain the highest statistical power/precision in answering the question about the best intervention for a given condition.

This project relates to the following EPSRC priority areas:
Data to knowledge
Mathematical analysis and its applications
New mathematics in biology and medicine
Statistics
Emergence and non-equilibrium physics

Publications

10 25 50
publication icon
Davies A (2021) Network meta-analysis and random walks

publication icon
Davies AL (2021) Degree irregularity and rank probability bias in network meta-analysis. in Research synthesis methods

Studentship Projects

Project Reference Relationship Related To Start End Student Name
EP/R513131/1 30/09/2018 29/09/2023
2091994 Studentship EP/R513131/1 30/09/2018 30/03/2022 Annabel Davies
 
Description We have shown that the structure of a network impacts the performance of network meta-analysis (a statistical procedure for combining data from many medical trials of more than two treatments). This work has been published.

We have made a novel interdisciplinary connection between NMA and an area of statistical physics (random walks). This work has been accepted for publication and is under production.

We have developed new models for survival analysis with time dependent covariates. We have applied these models to dynamic prediction and compared them with existing methods. We have submitted a manuscript for publication.

We are currently writing a review article to introduce network meta-analysis to statistical physicists. We also highlight existing and potential connections between the fields.
Exploitation Route The work on NMA structure can be used by medical researchers to assess potential biases that may occur due to network structure.

The connection between NMA and statistical physics provides new insight into the methodology. One of the methods developed by this approach has been integrated into commonly used medical statistics software. This produces more reliable results for clinicians and statisticians who use this software in medical research.

We hope that our survival analysis models will be used by medical researchers who would like to integrate time independent covariates into their survival analysis. Our models
come somewhere in between the two standard approaches in terms of modelling and computational complexity. Compared to joint models they provide a simpler, more streamlined and faster approach and rely on fewer modelling assumptions and parameters. Compared to landmarking they make use of more of the available data.
Sectors Healthcare

Pharmaceuticals and Medical Biotechnology
 
Description Work helps to improve statistical methods in medicine. Collaborations with medical statisticians is ongoing. We have improved statistical software used by medical researchers.
Sector Healthcare
 
Description Ton Coolen 
Organisation King's College London
Country United Kingdom 
Sector Academic/University 
PI Contribution We are working on a project on survival analysis. I have done the required derivations, performed simulations and analysed real data.
Collaborator Contribution The project idea was Ton's. He has been available in weekly meetings for project discussions.
Impact Biostatistics Statistical Physics
Start Year 2019
 
Description University of Frieburg 
Organisation Albert Ludwig University of Freiburg
Country Germany 
Sector Academic/University 
PI Contribution The project is about the connection between random walks and NMA. The project idea was my supervisors and we reached out the Freiburg group who are experts in NMA. I have carried out all analytical calculations, developed proofs and derivations, and performed simulations using my own codes.
Collaborator Contribution The collaborators from Freiburg have provided insight into network meta-analysis , have helped to place my work in the wider perspective, and have contributed to discussions about the project.
Impact Multi-disciplinary: Network meta-analysis (medical statistics) Random walks (statistical physics)
Start Year 2020
 
Title netmeta [Version 2.0-0] 
Description R package implementing frequentist methods for network meta analysis 
Type Of Technology New/Improved Technique/Technology 
Year Produced 2021 
Open Source License? Yes  
Impact Improved method for calculating the contribution matrix in network meta analysis based on random walk analogy 
URL https://CRAN.R-project.org/package=netmeta
 
Description School talk 
Form Of Engagement Activity A talk or presentation
Part Of Official Scheme? No
Geographic Reach Local
Primary Audience Schools
Results and Impact I recorded a talk to be shown at a local high school. This included an overview of my PhD and a discussion of Women in STEM. The aim was to show what careers can lead from studying STEM subjects and highlight issues of gender diversity and discuss why they exist. The talk had specific places to pause and allow discussion with their peers/teacher.
Year(s) Of Engagement Activity 2020