Seminars (IDP)

Abstract: Many COVID-19 research projects have elements of forecasting, planning and urgent decision-making that must be executed before solid data are available.  Some approaches, which neglect formal uncertainty quantification, can compromise estimates of low-probability/high-consequence outcomes: this could be a critical flaw when evaluating complex COVID-19 risks.  Adopting a structured elicitation procedure can reduce risk assessment weaknesses associated with inappropriate and sometimes misleading parameter and modelling assumptions. Several varieties of expert elicitation are practised, often categorised as 'behavioural aggregation' or 'mathematical aggregation'.  Cooke's Classical Model is of the latter class and we have used it extensively to provide scientific support for critical public safety decisions during volcanic eruptions. The method has garnered a track record for formalising the determination of input variable or parameter uncertainty distributions for risk modelling when conventional data do not exist or are undependable, e.g. with the adaptive Bayes Belief Net formalism for combining disparate strands of uncertain scientific evidence.  The underpinning algorithm provides empirical control on expert performance scoring and weighting for combining experts’ judgments into an auditable, rational consensus.  While it has become an established and validated elicitation procedure in many disciplines, this approach has not found widespread use in epidemiological modelling, even though it is well-suited to risk issues relating to emergent zoonoses.  We briefly summarise the principles of the method and how it is applied to real-world critical decision problems, mentioning a few case studies closely analogous to the present coronavirus problem.   Resources: The best link to materials relevant to my talk is Roger Cooke’s website: <rogermcooke.net> It has a lot of relevant stuff, and includes a link to a video of a 1-hour talk on the Classical Model for Structured Expert Judgment, which I gave at CDC Atlanta in May 2107.  

Abstract: The coronavirus disease 2019 (COVID-19) pandemic has triggered efforts by multiple modeling groups to forecast disease trajectory, assess interventions, and improve understanding of the pathogen. Such models can often differ substantially in their projections and recommendations, reflecting different policy assumptions and objectives, as well as scientific, logistical, and other uncertainty about biological and management processes. Disparate predictions during any outbreak can hinder intervention planning and response by policy-makers, who may instead choose to rely on single trusted sources of advice, or on consensus where it appears. Thus, valuable insights and information from other models may be overlooked, limiting the opportunity for decision-makers to account for risk and uncertainty and resulting in more lives lost or resources used than necessary. We advocate a more systematic approach, by merging two well-established research fields. The first element involves formal expert elicitation methods applied to multiple models to deliberately generate, retain, and synthesize valuable individual model ideas and share important insights during group discussions, while minimizing various cognitive biases. The second element uses a decision-theoretic framework to capture and account for within- and between-model uncertainty as we evaluate actions in a timely manner to achieve management objectives.   Resources: ·         Bjørnstad, O., Shea, K., Krzywinski, M. & Altman, N. (2020) Modelling Infectious Epidemics. Nature Methods (2020). https://doi.org/10.1038/s41592-020-0822-z ·         Associated shiny app: https://github.com/martinkrz/posepi1. ·         Outreach seminar “Disease outbreak control: Harnessing the power of multiple models to work smarter, not harder”: https://science.psu.edu/frontiers 

https://science.sciencemag.org/content/368/6491/577.summary

Abstract:  Doctors, lawyers, priests, teachers, politicians, fund managers, researchers, and more - how can we get experts to serve us? We have five main ways: pay for results, require procedures, or pick based on track records, prestige, or loyalty. Picking on prestige is most common. But paying for results seems most robust, though for cash-paid experts it requires that customers prevent expert coordination, or wait long to pay. For experts who forecast, paying for results can often be achieved via prediction markets, which have great untapped potential.


Resources: http://hanson.gmu.edu/expert.pdf http://hanson.gmu.edu/futarchy.html http://www.overcomingbias.com/2019/07/radical-pay-for-results.html http://www.overcomingbias.com/2020/05/five-ways-to-rate.html http://www.overcomingbias.com/2020/06/science-2-0.html  

Topic – Complex Models

The Royal Society initiative to provide Rapid Assistance in Modelling the Pandemic (RAMP) has garnered a large number of scientists across the UK working voluntarily in a wide range of issues associated with the current pandemic. In this talk I will describe the outputs of the fifty odd scientists with expertise, broadly speaking, in fluid mechanics who have been considering aspects of the airborne transmission of the virus. Post lockdown major risks of infection are associated with occupancy of public buildings such as restaurants, schools, and other places where people congregate indoors, and in public transportation. I will describe what we have learnt over the past couple of months about the impact of building ventilation, the deposition of droplets, modes of exhalation and inhalation and other factors such as the effects of occupancy levels and the movement of people in these indoor environments, and what seem to be the key outstanding questions.

During a pandemic like COVID-19, there is an urgent need to understand how easily the infection is spreading. Values like the reproduction number R, which measure transmission in a population, have therefore become prominent terms. But how are these numbers estimated? What are their limitations? And what can they tell us about how to control epidemics?

Jonty Carruthers, Martin Lopez-Garcia, Grant Lythe, Carmen Molina-Paris (Leeds). Joseph Gillard, Thomas R Laws, Roman Lukaszewski
(Dstl)
With a mouse infection model, agent-based computation and mathematical analysis, we study the pathogenesis of Francisella Tularensis
infection. A small initial number of bacteria enter host cells and proliferate inside them, eventually destroying the host cell and releasing numerous copies that infect other cells.  Our analysis of disease progression is based on a stochastic model of a population of infectious agents inside one host cell, extending the birth-and-death process by the occurrence of catastrophes: cell rupture events that affect all bacteria in a cell simultaneously.  We compare our analysis with the results of agent-based computation and, via Approximate Bayesian Computation, with experimental measurements carried out after of murine aerosol infection with the virulent SCHU S4 strain of the bacterium.
If I have time, I will also talk about Ebola, still a significant risk to humankind. Synthetic virology has been used to clone and manufacture two deletion defective genomes. These genomes were tested with Ebola virus using in vitro cell culture and shown to inhibit viral replication.  From in vitro experimental data, we identify parameters in a mathematical model of the infection. We examine the time an infected cell spends in the eclipse phase (the period between infection and the start of virus production), as well as the rate at which infectious virions lose infectivity.

The SARS-CoV-2 outbreak has infected over ten million, and killed over 500,000, individuals worldwide, making the development of antiviral treatments against this virus a priority. Recent studies have identified Remdesivir as an effective antiviral treatment option for COVID-19. I will present a new stochastic agent based model for the intracellular dynamics of a SARS-CoV-2 infection that includes details on all essential steps of a viral life cycle, from viral entry to virion release. Using this model, I evaluate the effect of Remdesivir on the production of new virions. In a next step, I propose an intercellular model to study the viral dynamics within the body of an infected patient. The intercellular model is fitted to data recording the progress of viral load in 12 patients with COVID-19, showing how differing strengths of immune response can explain the variety of infection spans seen in patients. Coupling the intra- and intercellular models allows for a comparative analysis of Remdesivir therapy for patients with different types of immune response.

Tuberculosis (TB) is an infectious disease caused by Mycobacterium tuberculosis. Despite significant recent advances, TB is the biggest infectious killer globally - someone dies from the disease every 18 seconds. I will describe a hybrid multiscale individual-based model that has been developed to study disease progression and treatment in the human lung. The model contains discrete agents which model the spatio-temporal interactions (migration, binding, killing etc.) of bacteria and immune cells. Chemokine and oxygen dynamics are also included, as well as Pharmacokinetic/Pharmacodynamic models, which are incorporated into the model via PDEs. In this work, I explore concepts of relapse and latent TB disease, and their effect on treatment outcome.

In order to estimate the patterns of movement, interaction with public spaces and potential risk of exposure to COVID-19, we need a detailed dataset containing individuals, their characteristics and there geographical location. These individuals are used in other models within the DECOVID ecosystem. We use spatial microsimulation techniques to generate the baseline population of individuals and an algorithm for assigning individuals to household to provide estimates at Middle Super Output area level. Further attributes are added to this population, derived from survey data using propensity score matching. We assess the spatial distribution of these populations and the daily activity spaces they inhabit.

Chair: Peter Challenor

Chair: Peter Challenor

Chair: Mark BirkinMeeting Host: WIll Taylor
Timetable
1:30 - 1:35 Introduction and overview of Urban Analytics. Mark Birkin1:35 - 1:45 Modelling journeys to work & business premises, Ying Jin
1:45 – 1.55 Spatial cooperation during a recurring disease outbreak, James Burridge1:55 – 2-05 Streetspace allocation analysis in London, Nicolas Palominos
2:05 – 2.15 Predictive modelling of social distancing and compromised distancing in determining railway passenger boarding rates and flows, David Fletcher 2:15 - 2:30 Discussion2-30 to 2-55 Break
2-55-3-00 Introduction to the Small Spaces Task, Mike BattyLonger Talks (~15 mins)
3-00 – 3.15 Fire Safety Centre/Centre for Numerical Modelling and Process Analysis, University of Greenwich, Ed Galea,
3-15 – 3-30 Network Rail, Nigel Best et al.,3- 30-3-45 Sainsbury’s, Dave Romano-Critchley et al.,3-45 – 4-00 Q and A 15 minutes with comfort breakLightening Talks (~10 mins)
4-00-4-10 CASA-UCL/Tesco, Fabian Ying,
 4-10-4-20 PWC AI Team, Technology & Investment Group,  Artem Parakhine/ Shabnam Rashtchi
4-20-4-30 CEGE, UCL, Thorsten Stoesser,
4-30-4-40 Martin Centre, U Cambridge, Ronita Bardhan
4-40 – 5pm Q and A for any topic of the small spaces task and any other in the afternoon

Chair: Peter Challenor

Chair: Peter Challenor

We developed a stochastic SEIR model for the COVID-19 epidemic at a fine spatial scale, using mobile phone mobility data, to describe the geographical spread of the virus. The model is developed for the very early phase of the epidemic, when movement of individuals plays a key role. In the beginning, importation of the virus into the region of interest is decisive, and we estimate the proportion of unknown imported cases. Our model represents non-pharmaceutical interventions to contain the epidemic by means of a regionally varying step function of the effective reproduction numbers. The regionally varying effective reproduction numbers are estimated by sequential Approximate Bayesian Computing, using hospitalisation data of the infected individuals at regional level. For prediction, we develop a way to regularise the mobility matrices, to conserve the geographical distribution of the population. This allows adequate long term predictions of all quantities of interest. Uncertainty in the parameters (both the estimated ones and the ones learned from the literature) is prolonged into the future by simulation. We use our model to describe the history of the first phase of the COVID-19 epidemic in Norway, during which social distancing and hygienic measures have been adopted together with teleworking and school closure and reopening. The result of these measures have reduced the presence of the virus in Norway to such a low level, leading to a relaxation of restrictions, to resemble the early phase of the epidemic, making our model again important. We compare the results of our model to the ones obtained by a similar nonregional model. We also developed a version of the model which has a time varying reproduction number. In this case we resort to Sequential Monte Carlo for inference. I will discuss the difficulties in making predictions using this model.  This is joint work with  Birgitte Freiesleben de Blasio, Solveig Engebretsen, Gunnar Isaksson Rø, Alfonso Diz-Lois Palomares, Kenth Engø-Monsen, Anja Bråthen Kristoffersen and Geir Storvik.

Speakers:
3:00 - 3:45 Matt Keeling   
3:45 - 4:30 Chris Jewell    
4:30 - 5   Discussion

1000-1030 Tim Lucas (Imperial College): Engagement and adherence trade-offs for SARS-CoV2 contact tracing
1030-1100 Chiara Polito (INSERM, Sorbonne Université) Anatomy of digital contact tracing: role of age, transmission setting, adoption and case detection
1100-1130 Michelle Kendall (University of Oxford): title tba

1000-1100: Modelling different strategies for reopening schools, the impact of test and trace interventions, and risk of occurrence of a secondary COVID-19  
                 epidemic wave in the UK

09:30-10:00: Ben Goldacre (University of Oxford) Secure and agile access to extended/linked NHS data
10:00-10:30 Nigel Shadbolt (Open Data Institute) Towards a national data infrastructure
10:30-11:00 Ulrich Paquet (DELVE) Data readiness in an emergency

Three  trends enhance the probability of global catastrophes ,    First, the rising global population, more demanding of energy and resources, leads to novel anthropogenic pressures on the biosphere -- climate change, loss of biodiversity, etc .    Second, the greater interconnectedness of our civilisation allows pandemics to rapidly cascade globally, and enhances our vulnerability to breakdown in supply chains, financial networks, etc .    Third, novel technologies -- bio, cyber and AI -- empower small groups with the ability (via error or terror) to cause massive (even global) disruption. Coping with this threat presents a challenge to governance: it will become ever harder to sustain the three goals of offering all citizens privacy, security and freedom.

In The Netherlands, the first wave of SARS-Cov-2 was controlled by what politicians like to call an “intelligent lockdown”, whereas the current wave is being addressed by a “semi-lockdown”. In this seminar, we’ll highlight and dissect the main components of the Dutch control strategy and will explore and compare the potential impact of some of these components using an individual-based model of COVID-19 transmission that included geographical stratification, individual heterogeneity in exposure and contribution transmission, and assortative mixing. In addition, we’ll explore the expected course and outcome of several controversial hypothetical strategies aimed at generating herd immunity though natural infection by a controlled release of the epidemic. Last, we’ll draw some parallels between said controversial strategies and what is currently happening in practice in The Netherlands.

Valarie Isham and Denis Mollison