Skip to main content

Our Statisticians contribute widely to conferences and publications to help enhance industry knowledge on statistical methodologies for clinical trials and broader aspects of healthcare.  In this Q&A, we talk with Natalie Thompson, a Statistician II at Veramed, about her work on statistical inference in modelling spatial epidemics, presented at the 2021 PSI conference.  The presentation was formed from her dissertation as a Masters’ student at the University of Lancaster. While her application in this instance was measles, the work is highly relevant in the context of COVID-19. In this interview, she shares her insights, and we offer her slides for those who may have missed the live presentation.

What statistical problem were you trying to address?

I decided to bring together the spatial epidemiology and MCMC modules of my MSc in statistics into a dissertation reviewing approaches to modelling the spread of infectious diseases. I chose to examine two previously used methods and applied them to specific disease data – in this case, measles.

Infectious diseases can be challenging to model, yet we can take steps to overcome these issues. For my dissertation, and in this presentation, I considered the Rhode Island Measles Epidemic and demonstrated how the Reed-Frost Model could be used to model infectious disease data.  Then, I examined another outbreak – the Hagelloch Measles Epidemic and considered how we could apply a Logit model to analyse this data.

What difficulties are there in modelling infectious diseases?

Missing data is a major cause of issues in infectious disease modelling.  In particular, it can be problematic to ascertain when an individual was initially exposed and infected, and this lack of information makes it difficult to determine the spread. However, Monte Carlo methods based upon Markov chains (MCMC methods), are invaluable tools to help overcome this problem.

What work did you undertake?

I decided to keep things relatively straightforward with my modelling choices. So, I excluded more complex temporal aspects at the outset (considering the impact of time on the spread of infection) on the assumption that children would be coming into contact with the same individuals daily at school.

The first model, under consideration, the Reed-Frost model, is visually simple to understand. However, while working with it, I found that it became complex and unwieldy very quickly. Despite its initial visual simplicity, it didn’t transfer particularly well onto a larger scale, even once you started examining households larger than three people. As a simple contact model, it was also unable to account for a range of influencing factors like age, sex or specific households. The impact of this complexity and rigidity is shown in my slides.

I then considered an alternative model – the Logit model – using a different set of epidemic data. For this work, I needed to do some initial exploratory data analysis to inform the factors for inclusion. Based on that exploration, we decided that household, class and spatial distance were the main factors influencing the measles spread. I ultimately found the Logit model more flexible to work with and well suited to the task at hand.

What does this piece of work add to the existing consensus?

I felt the dimension of public health statistics was particularly relevant and timely. Even though we may not work in this field daily, it is helpful to understand the challenges healthcare colleagues face.

By comparing two different, yet straightforward ways of modelling epidemic data, we can give colleagues insight into practical options we could apply to other diseases. While the Reed-Frost model was probably too limited to translate to any substantial epidemic setting, the Logit model could feasibly be extended into larger outbreaks of other diseases if informed with some exploratory data analysis.

 For more detail about the modelling approaches Natalie used, and the results, click below to download the presentation slide deck.

About Natalie Thompson

Natalie Thompson is a Statistician II at Veramed. She joined us in 2019 having completed her MSc in Statistics with a medical pathway at Lancaster University. She also holds a BSc in Mathematics from Loughborough University.

Natalie Thompson