Seminars (IDP)

Abstract:

The incorporation of uncertainty in assessments and
predictions from mathematical models is critical, especially if the models are
to be used to support real-world decisions. In fast moving situations such as a
global pandemic of an infectious disease, then data to parameterise models are
typically patchy and incomplete, and sufficient suitable data may not exist for
many parameters in model runs of possible future scenarios. In such cases
expert judgements can play an important role, both to specify uncertainty
distributions for parameters with no available data and to supplement data
where they are available (via Bayes Theorem, or more informally). In this talk
I will discuss the elicitation of uncertainty distributions for individual
unknowns from a single expert, the combination of the opinions of multiple
experts on an unknown into a single uncertainty distribution and the
elicitation of graphical models, with an emphasis on Bayesian networks, to
produce suitable model structures from experts over multiple dependent
unknowns. I will emphasise a behavioural aggregation approach, the SHeffield
Elicitation Framework (SHELF), for the combination of the opinions of multiple
experts, which will complement the talk from Prof Aspinall on a mathematical
aggregation approach, the Classical Method. A running example on the
development of a diagnostic test will be used to illustrate the ideas, and I
will try to bring out particular issues surrounding infectious disease
modelling.



Resources:

·A probabilistic judgements e-learning course, aimed at
explaining elicitation generally and the Sheffield ELicitation Framework
specifically:http://www.tonyohagan.co.uk/shelf/ecourse.htm

·A textbook providing comprehensive coverage of
elicitation: O'Hagan et al (2006). Uncertain Judgements: Eliciting Experts'
Probabilities, Wiley.

·Resources to conduct an elicitation using SHELF
including slide sets, advice and document templates:http://www.tonyohagan.co.uk/shelf/

·A series of short videos for an online course on
Structured Expert Judgment provided by TU Delft (links are near the top of the
page):http://rogermcooke.net/

·A textbook discussing the elicitation of probabilistic
models: J. Q. Smith (2010). Bayesian decision analysis: principles and
practice. Cambridge University Press.

Abstract: Expert
judgment can support policy decision-making in situations where data are
scarce, knowledge is incomplete and decisions are imminent. Research has demonstrated
both the potential value and the potential dangers of expert judgments, and
established the benefit of using systematic, structured procedures to elicit
judgments from experts which anticipate and correct for the most severe and
predictable limitations of expert opinion. In this presentation I outline the
origins and development of the IDEA protocol, a Delphi-style structured
elicitation procedure that combines psychologically robust interactions among
experts with mathematical aggregation of individual estimates and structured
question formats to improve the accuracy of elicited judgments. I discuss
examples and emerging insights from its use to support decision-making across a
range of applications including biosecurity, natural resource management and
real-time geopolitical forecasting. Drawing from these experiences I reflect on
the challenges presented by COVID-19 and how structured elicitation procedures
might best support epidemiological modelling and simulation efforts in
providing timely public health policy advice.



Resources:

IDEA protocol for structured expert elicitation
– key reference

1. Hemming,
Victoria, et al. "A practical guide to structured expert elicitation using
the IDEA protocol."Methods in Ecology and Evolution9.1
(2018): 169-180.

[This article provides a practical step-by-step
guide to carrying out a structured elicitation using the IDEA protocol. The
supplementary information includes ready-made templates and resources for
planning and implementing a structured elicitation using IDEA]

An excellent introductory blog post discussing
this article and the need for structured expert elicitation is available here: https://methodsblog.com/2018/03/27/idea-protocol-2/

Expert elicitation references – introductory
& overview

2. Morgan, M.
Granger. "Use (and abuse) of expert elicitation in support of decision
making for public policy."Proceedings of the National academy of
Sciences111.20 (2014): 7176-7184.

[Excellent overview article for using expert
judgment that is accessible for most audiences, and with an excellent reference
list for further reading]

3. Alahmadi, Amani, et al. "Influencing
public health policy with data-informed mathematical models of infectious
diseases: Recent developments and new challenges."Epidemics(2020):
100393.

[Includes coverage of the role and challenges for
expert elicitation from the perspective of modelling infectious diseases]



Cognitive Biases

4. Cognitive biases infographic. Structured elicitation
protocols include steps that attempt to reduce the impacts of many of these
biases on the judgments elicited from experts.

https://www.visualcapitalist.com/every-single-cognitive-bias/


A
more complete version of the cognitive biases infographic that includes
definitions of each of the biases is also available here

Topic – Complex Models

Topic – Dynamics of Transmission

Chair:
Peter Challenor Why do uncertainty quantification
Speakers;
Evan Baker (Exeter) Emulating Stochastic Models
Building emulators for complex models typically involves Gaussian processes. For stochastic models, the flexibility of a Gaussian process is a nice feature, but modifications are needed to account for the noisiness of simulations. In this talk I will summarise some key attributes of stochastic models and how these can change the emulation methodology. Additionally, I will briefly talk about the simulation design issues that arise for stochastic models.
Jeremy Oakley (Sheffield)  - Introduction to Probabilistic Sensitivity Analysis
Mathematical models of infectious diseases invariably have uncertainty about the correct values of some of their model inputs/parameters. This induces uncertainty in the model outputs. In some situations, it may be desirable to reduce this uncertainty, by collecting more data about uncertain model inputs, before using the model outputs to inform decisions. However, it is unlikely that all inputs are 'equally important': some will contribute to output uncertainty more than others. I will discuss how probabilistic sensitivity analysis can be used to identify which uncertain inputs are most influential, and describe simple computational tools that can be used for implementing the analysis, based on a random sample of model runs.

 

Over the past years oncolytic viruses have generated much interest in cancer therapy, mainly due to the fact that once a virus is injected into the patient it can actively search for cancer cells and destroy them. However, the anti-tumour effect of oncolytic viruses is greatly diminished by anti-viral immune responses (both innate and adaptive responses), as well as by physical barriers inside the tumour. 
Using various single-scale and multi-scale mathematical modelling approaches, we will investigate the delicate balance between anti-viral and anti-tumour immune responses in the context of virus-tumour-immune interactions.

Chair: Mark Birkin

9:30 - 11:30 - Session One:Urban Analytics
9:30 - 9:50 Introduction and overview, Mark Birkin
9:50 - 10:10 Microsimulation, Nik Lomax, Karyn Morrissey and Jamil Nur 
10:10 - 10:30 Modelling behaviour, Adam Dennett, Richard Milton, Ying Jin and Mike Batty 
10:30 - 10:50 Epidemic simulation and scenario planning: Gavin Shaddick, Nick Malleson

10:50 - 11:30 Discussion

11:30-12:00 - Break
12:00 - 13:00 - Session Two:Small spaces 
12:00 - 12:40 - Introduction to key projects: Mike Batty, Gerry Casey, Ed Galea & others

12:40 - 13:00 - Discussion

Interaction based risk - an agent based analysis of London   We have seen unprecedented changes in transport demand and supply, as a result of Covid-19. Transportation can play a key role in the regional economic recovery by enabling people to safely access work, healthcare, education. But we know that Covid-19 is highly contagious, and there are no current vaccines. Transport agencies will need to plan safe services that minimise crowding, maintain critical service schedules, and mitigate new challenges as operating resources are pinched.   Established transportation modelling techniques are not designed to analyse how individuals travel, at a granular level. This makes it challenging to estimate how changes in transport demand (the result of physical distancing policies) may impact public transport and identify where coronavirus transmission risk may be greatest. Agent based models (ABMs) provide granular insight into how individuals behave. The aggregate consequences of individual behaviours can be analysed to understand where and when congestion may happen, which populations make use of which public transit services, and other network-wide consequences.   Arup has been working with TfL to understand how public transport may impact the risk of Covid-19 transmission in London. We have extensive experience in developing ABMs – including an ABM for London, developed as part of a previous, collaborative TfL-Arup research initiative. We have been testing a new epidemiological model in conjunction with our London mobility ABM that attempts to take into account "virus infection dynamics" (EpiSim , developed by Technische Universität Berlin).

Chair: Peter Challenor

Chair: Peter Challenor

Chair: Peter Challenor

Chair: Peter Challenor

Speakers:
3:00 - 3:45 Glenn Marion   
3:45 - 4:15 Martina Morris   
4:15 - 5:00 Thomas House   
5 - 5:30 Discussion

1500-1530: Maryam Aliee (Warwick) Estimating the time to extinction of infectious diseases in mean-field approaches

1500-1530: tba
1530-1600: tba
1600-1630 Julia Lane (NYU) State of the data – a collaborative perspective
1630-1700 Sarah Henry (ONS) State of the data – a national statistics perspective

Zoonotic diseases have long been a major burden on human societies and are expected to increase in frequency and impact as we interact more with the animal world and as the global population increases in both size and productivity. Fortunately, new genomic tools, particularly metagenomic next-generation sequencing (mNGS), provide a uniquely powerful means to rapidly reveal the microbial composition of any sample without bias, provide key information on the diversity, structure and evolution of the virosphere, help determine how microbes move across the human-animal interface and the drivers of disease emergence, and reveal the origins of specific epidemics. Herein, I demonstrate the utility of mNGS for pathogen discovery and understanding disease emergence on clinically actionable time-scales. In doing so, I will demonstrate how these genomic tools can form a key component to new approaches to pandemic preparedness. As a case study will focus on the initial emergence of COVID-19 (SARS-CoV-2) at the end of 2019. I will discuss the most likely theories for its origin and emergence, and consider why coronaviruses seem particularly able to jump species boundaries and emerge in new hosts. I will conclude by outlining the ways in which we can potentially prevent pandemics like that of COVID-19 ever happening again.

Coivd-19’s arrival in the human population
was inevitable. There is a huge
diversity of viral pathogens circulating in bats and other small mammals. Three groups of people are exposed to them
through their livelihoods: traders in the wildlife trade, the miners and
loggers destroying tropical forests and those working in intensive
agriculture. The initial dynamics of
novel virus in these three groups of people and their families determine
whether novel viruses will spread into urban areas and from there to the rest
of the world.

This talk will fall into three sections:
(1) Initially I’ll discuss ways to estimate the diversity of viruses with
zoonotic potential and how this determines the risk they will spread from the
initial crossover hosts into the rest of the human population. (2) I’ll then briefly discuss some earlier
models for how forest destruction changes the risk of transmission of viruses
from forest species to those converting the forest or those living in the newly
converted agricultural matrix. (3) In
the final section, I’ll develop some economic approaches that compare the cost
of modifying the activities that increase risk of viral emergence with the
current estimated cost of the Covid19 pandemic.

Qatar has experienced one of the most intense SARS-CoV-2 epidemics with >55,000 laboratory-confirmed infections per million population. The country’s robust investment in epidemiology research and primary data collection allowed the conduct of a series of studies that yielded novel and influential data on this infection. In this presentation, an overview of the insights learned about SARS-CoV-2 epidemiology will be provided with a focus on epidemic dynamics, reinfection, infection severity and fatality rates, and herd immunity.