Introduction

Statistical and mathematical models are a crucial component in understanding the state and evolution of the COVID-19 pandemic, and for assisting decision makers with the evidence needed to:

  • inform and support local and national actions to respond effectively to actual or expected increases in infection levels;
  • efficiently manage and deploy resources across the NHS Test and Trace programme, for example to prioritise testing capacity;
  • effectively target communications and non-pharmaceutical interventions to local areas expected to be at increased risk of high incidence and prevalence.

The Alan Turing Institute and the Royal Statistical Society (RSS) are partnering on this programme of work to support the Joint Biosecurity Centre (JBC), which is part of NHS Test and Trace, based at the Department for Health and Social Care, by providing an independent source of statistical and mathematical modelling and machine learning expertise to address policy-relevant research questions. The Turing-RSS Lab has been established for one year in the first instance, and will work with JBC to identify and prioritise projects that require reflective thinking to develop and evaluate a solution, but are sufficiently focused so that useful outputs can be obtained within months, thereby filling a current gap between rapid-response consultancy and most existing research funding streams.

Goals

The Turing-RSS Lab focuses on co-developed research projects which both meet a need for JBC and the need for rigorous research which can add value to the existing body of knowledge.

  • Provide independent, rigorous modelling and analysis to deliver new insights in the evolving fight against COVID-19.
  • Provide further understanding of COVID-19 to the public and wider scientific community.
  • Enhance capacity within JBC to better forecast and model the current and future epidemics.

Building a community

Community-building connects external expertise with the COVID-19 data science and analytics that takes place in JBC, NHS Test and Trace, and Public Health England (PHE). Activities will identify key and cutting-edge models and analytics, support academic input and discussion between different producer and user communities, and help visibility and understanding of the contribution of data science outside and inside government, addressing the evolving picture of COVID-19.

 

Working values

The focus of the Turing-RSS Lab is on supporting the government to undertake responsible and risk-aware design, development and deployment of statistical and mathematical modelling and machine learning.

The Turing-RSS Lab believes that critical assessment of its work by the research community and the public at large will help to improve the quality of any advice that it provides. Critical assessment through open science is an important tenet of this collaboration.

The Turing and RSS are able to draw upon a wide range of expertise through open call, to ensure recruitment of the most talented data scientists, with diverse backgrounds and experience.

All algorithms are to be designed and developed in a transparent and reproducible manner and delivered with sufficient detail that external research teams can replicate results if they have access to the dataset. This requires:

  • sharing the algorithms and methods publicly, so that anyone may examine them and share their insights and input with the Institute and RSS;
  • speaking openly in relation to our work, in line with the Turing’s status as an Independent Research Organisation;
  • adhering to ethical research principles, worthy of public trust, justifiable, fair and non-discriminatory.

Current projects

Nowcasting positive test counts with reporting lag

Research team

Dr Radka Jersakova, The Alan Turing Institute
Dr James Lomax, National Cyber Security Centre and The Alan Turing Institute

To monitor the current state of COVID-19, the UK government tracks the number of positive tests in each local authority. Since it takes time to process PCR swab tests, there is a delay of up to five days before all positive test results are reported.

The goal of this project is to 'nowcast' the number of daily positive test counts up to the present date. A 'nowcast' is a prediction informed by analysis of data currently available. Using statistical models we can infer the expected final count using the incomplete data as it arrives. This estimate can be used to make up for the lag in data reporting and aid decision making.

Timeframe

From August 2020

Spatial and temporal modelling of incidence and prevalence of COVID-19

PI

Professor Marta Blangiardo, Imperial College London

Research team

Tullia Padellini, Imperial College London
Annie Mallon, MRC Harwell
Ruairidh King, MRC Harwell 
Luis Santos, MRC Harwell

Our goal is to estimate the prevalence of COVID-19, combining several sources of data, accounting for their biases and uncertainty. 

We aim at predicting the burden of the disease by integrating two different type of information on the number of cases: direct estimates (such as randomized surveys and testing programs) and indirect estimates (such as hospital admissions). We provide a flexible modelling framework, which is adjusted for known risk factors and accounts for spatial as well as temporal dependencies in our data. 

Timeframe

From November 2020

Estimating COVID-19 prevalence and transmission from multiple sources: de-biasing Pillar 2 data

Research team

Brieuc Lehmann, University of Oxford
George Nicholson, University of Oxford
Annie Mallon, MRC Harwell
Ruairidh King, MRC Harwell
Luis Santos, MRC Harwell

Our goal is to estimate COVID-19 prevalence and transmission rates at a fine-scale level, such as local authority, by harnessing data from multiple testing sources. We are designing a statistical model to adaptively adjust for biases and coherently combine information across multiple data streams.

Background

The daily or weekly number of positive COVID-19 tests in a region is widely used as a proxy for the local number of infected individuals. 

Multiple testing sources

Positive test numbers arise from: 

  • Randomized surveillance (REACT study, ONS survey).
  • Pillar 2 testing focused on testing symptomatic individuals.
  • Local mass testing at the level of cities, universities, care homes etc.  

Testing bias

Tests results are subject to sampling and operational influences:

  • Ascertainment bias: symptomatic individuals are prioritized for testing, so the rate of positive tests is greater than the actual disease prevalence in the population.
  • False positive/negative test results: tests for COVID-19 infection, such as PCR and lateral flow, vary in sensitivity and specificity.
  • Weekday effects: the numbers of tests performed depends strongly on the day of the week.

Timeframe

From November 2020

Effectiveness of Non-Pharmaceutical Interventions (NPIs)

This project will seek to understand the effect of various non-pharmaceutical interventions on a range of outcomes which will be specified as the project executes. The ultimate research goal is to provide a deep understanding of the menu of policy options available to the government to help stop the spread of the virus.

Read more about this project. Please note that applications for the open call have now closed.

Find out more

If you have any questions about this work, please contact [email protected].