RAPID ASSISTANCE IN MODELLING THE PANDEMIC: RAMP

A call for assistance, addressed to the scientific modelling community

Coordinated by the Royal Society

RAMP update, 8 May

This update informs the RAMP community about what has happened in the three weeks since the previous (16 April) update – which remains available. We do not repeat what we said on 16 April, except to reiterate that he RAMP team remains immensely grateful for all the support offers received.

The previous update includes advice for those who have not yet been tasked with specific roles to proceed by yourselves if you prefer. If you tell us what you are planning, and if we think it worth adopting as an official RAMP project, you may then be able to request additional RAMP volunteers for your team.

Progress since 16 April

1. We have continued to reinforce SPI-M teams with skilled volunteers. This effort has been appreciated and has facilitated research outcomes reported to SAGE.

2. In partnership with RAMP, an online programme of the Isaac Newton Institute on Infectious Dynamics of Pandemics has been established to run 5 May - 31 August. Prof Deirdre Hollingsworth chairs the Organizing Committee. For details see https://www.newton.ac.uk/event/idp. This programme will span invitation-only technical discussion sessions, via workshops open for qualified applicants, to the production of online pedagogical materials such as those already available (from a previous programme) at https://www.newton.ac.uk/event/idd/seminars

3. The RAMP discussion forums are now open at https://ramp-forums.epcc.ed.ac.uk. All RAMP volunteers and their team-members are invited to sign up for the forums. In the event of heavy traffic, it may take some time to approve your registration. Of special significance is the Research Outputs: Community Review forum. There, you and your colleagues are invited to nominate, review and discuss research outputs of potential significance to managing the pandemic. A team of moderators, led by staff of the Mathematics and Biology Departments of the University of York, will use these "community reviews" to identify work potentially worthy of the attention of SPI-M and/or SAGE. Such work will be fed via a small triage team into the RAMP Rapid Review Group (see next item) for consideration. Outputs for community review can include datasets and codebases as well as preprints or journal papers. They can come from any source, but must be open access.

4. The RAMP Rapid Review Group (RRG) is led by Alain Goriely and Phillip Maini from the University of Oxford. The group is tasked to commission rapid expert assessment of research ouputs, reports and codebases nominated directly by SPI-M members, SAGE members, or RAMP Task Team Leaders, and decide whether these have the combination of scientific importance and policy relevance that merits their consideration by SPI-M /SAGE in their roles as advice channels to the Government.

5. Updates from the RAMP Task Teams:

5.1. The RAMP Task Team on new epidemic modelling is being led by Prof Graeme Ackland (Edinburgh). Developments include:

(a) a major consortium is being led by SCRC (Scottish COVID-19 Research Consortium), initiated by scientists from four core institutions in Scotland, but now has over 100 contributors from 12 organisations across the UK. This consortium is working on several separate codebases, mostly originating from prior work on animal epidemics.

(b) A team of RAMP volunteers from Cambridge has created a Python library for general compartment models, for example SEI(...)R, with unlimited disease stages, structured by age. The library simulates both stochastic and deterministic versions of the compartment dynamics. The library enables fully Bayesian forecasts, including data and parameter uncertainties convolved with intrinsic stochasticity. The library allows Bayesian estimates of NPI impacts, and can compute optimal strategies for these, given cost functions. A beta version of the library is freely available at https://github.com/rajeshrinet/pyross and has numerous Jupyter notebook examples. The team also has a geographical version of this code which will be released soon.

(c) A team from Nottingham has started work on developing simple, yet sufficiently realistic models, to be fitted to the limited data available to address questions such as how infection rates have varied over time (before and after the lockdown), quantify uncertainty about model parameters and predictions, and determine the ability of those models to capture features of the more complex, individual-based models.

(d) A team from Frazer-Nash Consultancy is developing a holistic cost model that relates intervention policy to health and economic impacts, accounting for the impact of public perception on policy effectiveness. The modelling approach, which has been used successfully in the energy and defence sectors, explicitly considers the uncertainty in future predictions and how this can be reduced through testing and data collection.

(e) A RED team of RAMP volunteers from Edinburgh has helped stress-test COVID-sim, the Microsoft implementation of the Ferguson/Imperial College model. The latter implementation is now open access at https://github.com/mrc-ide/covid-sim and has been ported to major UKRI computing platforms. With help from the RED team, an independent RAMP team from UCL has begun systematic parameter-sensitivity testing for COVID-sim.

5.2. The RAMP Task Team on urban analytics / social modelling has been set up in partnership with the Alan Turing Institute and is led by Prof Mark Birkin (Leeds & Turing), comprising about 25 RAMP volunteers. The team is working to address the following two goals: (a) to use modelling of human behaviour including use of transport systems, shopping and leisure facilities to better inform the parameters of epidemic models by understanding human encounter dynamics in various environments; (b) to integrate disease status, as an attribute, into models of human behaviour so as to create a new platform for epidemic modelling in which full social interactions are present from the outset rather than introduced to estimate parameters such as rates for infection.

5.3. A RAMP Task Team has been established to apply social modelling methodologies to human dynamics in small spaces (relative to those traditionally considered in urban analytics – for instance, within a supermarket). This team is led by Prof Mike Batty (UCL), and has about 20 FTE of staff effort from RAMP volunteers; alongside academics it includes staff from at least six partners across the commercial and industrial sectors.

5.4. A RAMP Task Team has been established to investigate and model environmental and aerosol transmission, including interaction of aerosols and droplets with air flow especially in buildings and other confined spaces, and its relation to the physical transport of virus particles between individuals either directly airborne or via intervening surfaces. This team is led by Prof Paul Linden (Cambridge) and Prof Chris Paine (ICL).

5.5. A RAMP Task team has been established to investigate and model the within-host disease dynamics of COVID-19 within individuals carrying the infection, to understand better the time evolution of the disease, its level of variability, possible groups at risk of severe outcomes or of being more infectious than others, and other factors. This team is led by Prof Mark Chaplain (St Andrews).

5.6. The Institute for Actuaries' Covid-19 Actuarial Response group, led by Mohammad Khan and Gavin McInally, has convened a RAMP Task Team to attempt to identify comorbidity factors, including but not limited to pre-existing medical conditions, and estimate the prevalences of these factors and combinations thereof across the population, with the goal of better informing decisions about shielding strategies during the unlock period and their consequences.

5.7. A RAMP Task Team, led by Steve Sparks and Willy Aspinall from Bristol, has been established to consider, on behalf of RAMP and DELVE (our sister-project on data analytics also run by the Royal Society), the potential role of Structured Expert Judgement (SEJ) for parameter estimation and decision support – such as the effect of opening schools on transmission rates – when data is of insufficient quality or quantity for fully evidence-based scientific advice. Two demonstration applications are being developed.

Ongoing developments: Data

Patient Data

6. Access to patient data, even anonymised, presents a serious bottleneck to epidemic modellers, including SPI-M teams, but particularly for those entering the field for the first time. A number of agencies across Government are now working hard to solve this problem and provide streamlined access. RAMP is in direct discussions with several of these organizations. However RAMP modelling teams cannot expect to receive large amounts of clean UK data on the time-scale they might like and even then such data may require permissions protocols.

7. One significant data access channel is HDRUK. This has a newly developed portal at https://www.hdruk.ac.uk/covid-19/, and RAMP is trying to take a generic request for stratified population data through the process and will follow up with other requests. Such data will be shared across RAMP to whatever extent the data-release conditions permit this. We are engaging with HDRUK to streamline and document the process and requirements for access to data that can then be used for RAMP research. RAMP volunteers are of course free to use this and other data-access routes directly for themselves, but must not claim without authorization to be doing so on behalf of RAMP.

8. With patient-specific data rather than stratified population data, the situation is more complex still. Generally it may prove necessary to import the model into a safe-haven where the data resides and export it again with parameters fit to that data but without the data itself. Such safe-havens so far exist separately in different UK Nations and data is not easily accessed without detailed permissions and not easily transferred from one safe-haven to another.

9. A possible route to obtain access to patient data, both population-level and individual, is for researchers or teams to find collaborators who are running closely related projects that can be amended to include the planned COVID-19 research. This would require the amendment practice for the original research to be followed. See for instance https://www.hra.nhs.uk/covid-19-research/covid-19-guidance-sponsors-site..., but note that procedures may differ between safe-havens.

10. Before patient data arrives, anyone with a new epidemic model (model A) is advised to test it first on synthetic data from one or more other models of similar or greater complexity (say model B). This allows the predictive power of model A to be ascertained in the context of a 'alternative pandemic' where model B is in fact true. If this predictive power is lacking, then it is likely also lacking for the true pandemic – unless the real-world evidence base for the prior assumptions behind model A is vastly stronger than the corresponding evidence base for model B. It may also be possible to test your model against real-world data from other countries that are further along the unlock path than the UK and/or have a more relaxed attitude to patient data. Such tests might significantly improve the prioritization of any request you make (via RAMP or otherwise) for UK patient data. They are also separately valuable for studies of how the UK response differs from that of other countries.

11. Anyone coming across datasets for patient outcomes that are clearly in the public domain and do not carry confidentiality issues, is warmly encouraged to post links to these resources on the RAMP forums under 'Research Outputs: Community Review' tagging the output as 'data' (as well as by subject area). Others are then invited to comment on their utility.

Social data

12. 'Social data' refers to data about human behaviour – typically mobility, commuting, shopping habits and the like. Large amounts of such data are already available, and the RAMP Task Team working on urban analytics / social modelling has strong experience in accessing it. Unlike patient data, this is 'big data' for which appropriate skills are needed. Commercial confidentiality restrictions very often apply but there may be ways to share such data with other teams across RAMP if a good case can be made for it. Additionally, RAMP (alongside DELVE) is actively investigating data-sharing arrangements with potential partners particularly in the mobile phone and banking sectors.

13. Anyone coming across datasets for social data that are officially in the public domain and do not carry confidentiality issues, is warmly encouraged to post links to these resources on the RAMP forums under 'Research Outputs: Community Review' tagging the output as 'data' (as well as by subject area). Others are then invited to comment on their utility.

Original call | Previous updates | Privacy Statement for the RAMP initiative | Contact