Impact case study database
Supporting decision making in West London NHS trusts during the COVID-19 pandemic.
1. Summary of the impact
Brunel’s research was incorporated into a toolkit that directly informed planning decisions for 2
NHS trusts (LNWH and Hillingdon), including 6 hospitals serving 1,290,000 citizens. LNWH used the simulations to plan COVID-19 capacity and service recovery, informing the pace and scale of health service provision and capacity expansion in response to pandemic waves, benefitting approximately 4,000 COVID-19 patients and improving the well-being and working conditions for approximately 2,600 hospital staff. This led to the trusts being able to treat 2.3 times more COVID-19 patients in their hospitals at the end of 2020, compared to the first wave capacity, and helped prevent medical services in West London being overwhelmed. The toolkit was also applied to analyse CovidSim on behalf of the Royal Society, with the findings contributing towards a shift by SAGE away from reliance on single codes.
2. Underpinning research
Brunel researchers created an epidemiological simulation toolkit for large-scale COVID-19
forecast analysis. The toolkit relies on a combination of agent-based modelling algorithms to
reproduce individual behaviours, ensemble simulations to efficiently account for statistical noise
and counterfactuals and automated access to large supercomputers (such as the Eagle
supercomputer in Poznan (PL), and the SuperMUC-NG computer in Garching (DE)) to facilitate
the rapid and reliable execution of these complex workflows. The research work performed to
establish the toolkit comprises two main themes: one concerning the Flu And Coronavirus
Simulator (FACS) and one concerning the VECMA toolkit for facilitating robust and reproducible
simulations on the high end computing scale.
The FACS code [REF1] has been developed from the ground up, and uses an agent-based
modelling algorithm with people represented as individual agents in conjunction with local
mapping data, demographic information and disease information. FACS serves to forecast the spread of Covid-19 in a local environment, such as a borough in London or a different city or region in the UK. The code has been validated against hospitalisation data of the first wave of COVID-19 from two NHS Trusts: London North West University Healthcare (LNWH) and Hillingdon Hospitals, and has shown good agreement with that validation data. The code also supports a wide range of public health interventions, such as social distancing, mask wearing and the closing of locations by type. To run FACS robustly, we perform ensemble simulations that cater for a large number of intervention scenarios, while also testing the sensitivity of key assumptions in specific forecasts, and accounting for the aleatory uncertainty in the code [REF6]. The FACS code has been developed in collaboration with the HiDALGO Centre of Excellence project (https://hidalgo\-project.eu\) and is currently applied to simulate COVID-19 spread in Madrid as part of that project.
The VECMA toolkit [REF2], co-developed by Groen and the VECMA Consortium, facilitates rapid and robust verification, validation, uncertainty quantification and sensitivity analysis for a wide range of scientific applications. It is used by researchers across 8 different application domains, and it enables the team at Brunel to quickly run large ensembles of simulations using remote supercomputers (using the FabSim component **[REF4]**), and enables them to rapidly identify key uncertainties in the simulations, and parameters to which the simulations outcomes are most sensitive (using the EasyVVUQ component **[REF3]**). We also used the ensemble execution approach in VECMAtk to analyse the sensitivities and limitations of the CovidSim code [REF5], which has been consulted for policy decision making by the UK government through the Scientific Advisory Group for Emergencies (SAGE). We found that the code derives 70% of its sensitivity from only 4 out of the 940 parameters and that the model’s validation errors cannot be fully attributed, in a probabilistic sense, to lack of knowledge about the correct values of these 940 parameters. We highlight important omitted factors, such as the modelling of care homes and medical facilities, and the effect of using face masks, that could be the reason for this discrepancy.
3. References to the research
REF1: Mahmood, I., Arabnejad, H., Suleimenova, D., Sassoon, I., Marshan, A., Serrano, A.,
Louvieris, P., Anagnostou, A., Taylor, S., Bell, D. and Groen, D., 2020. FACS: A geospatial
agent-based simulator for analyzing COVID-19 spread and public health measures on local
regions. Journal of Simulation, pp. 1-19, doi:10.1080/17477778.2020.1800422.
REF2: Groen, D., Richardson, R.A., Wright, D.W., Jancauskas, V., Sinclair, R., Karlshoefer, P.,
Vassaux, M., Arabnejad, H., Piontek, T., Kopta, P. and Bosak, B., 2019, June. Introducing
VECMAtk - Verification, Validation and Uncertainty Quantification for Multiscale and HPC
Simulations. In International Conference on Computational Science (pp. 479-492). Springer,
Cham. doi:10.1007/978-3-030-22747-0_36.
REF3: Wright, D.W., Richardson, R.A., Edeling, W., Lakhlili, J., Sinclair, R.C., Jancauskas, V.,
Suleimenova, D., Bosak, B., Kulczewski, M., Piontek, T., Kopta, P., Chirca, I., Arabnejad,
H., Luk, O.O., Hoenen, O., Węglarz, J., Crommelin, D., Groen, D. and Coveney, P.V.
(2020), Building Confidence in Simulation: Applications of EasyVVUQ. Adv. Theory Simul..
doi:10.1002/adts.201900246.
REF4: Groen, D., Bhati, A.P., Suter, J., Hetherington, J., Zasada, S.J. and Coveney, P.V., 2016. FabSim: facilitating computational research through automation on large-scale and
distributed e-infrastructures. Computer Physics Communications, 207, pp.375-385. Doi:
10.1016/j.cpc.2016.05.020.
REF5: Edeling, W., Arabnejad, H., Sinclair, R.C., Suleimenova, D., Gopalakrishnan, K., Bosak,
B., Groen, D., Mahmood, I. Crommelin, D., Coveney, P.V..The Impact of Uncertainty on
Predictions of the CovidSim Epidemiological Code. Nature Computational Science.
doi:10.21203/rs.3.rs-82122/v3.
REF6: Groen, D., Arabnejad, H. et al. VECMAtk: A Scalable Verification, Validation and
Uncertainty Quantification toolkit for Scientific Simulations. Phil. Trans. R. Soc. A (in press,
preprint available at: https://arxiv.org/abs/2010.03923\).
GRANT 1: Dr. Derek Groen, Verified Exascale Computing for Multiscale Simulations, European Commission Horizon 2020 Programme, June 2018 - June2021, EUR4,000,000 (EUR429,000 allocated to Brunel, equivalent to GBP379,643 [01-2021]).
GRANT 2: Dr. Derek Groen, HPC and Big Data Technologies for Global Systems, European Commission Horizon 2020 Programme, December 2018 - November 2021, EUR8,000,000 (EUR647,000 allocated to Brunel, equivalent to GBP572,562 [01-2021])
4. Details of the impact
NHS planning is typically driven by a mix of historical data and trend analysis. COVID-19 (a
novel virus), lacking such data, required rapid evidence analysis, synthesis, and simulation to
provide actionable local insight. Senior NHS managers (CIOs and direct reports) from West
London with Bell planned a collaboration strategy in March 2020. Brunel was then able to rapidly assemble academic teams deploying their research on evidence sifting (for model inputs), COVID-19 transmission and bed utilisation modelling. Research outputs were subsequently disseminated to the NHS through the delivery, presentation and discussion of 15 written forecast reports between April and December 2020, directly feeding into operational planning decisions of 2 NHS trusts, the London North West University Healthcare (LNWH) Trust and the Hillingdon Hospitals (HH) Trust, which together are responsible for the healthcare provision for 1,290,000 citizens. Scenarios were co-designed with the NHS as research was developed and validated. Forecasts were then disseminated to NHS analytics groups to further support short-term, elective and social care planning and transformation.
The epidemiological simulation toolkit has been used repeatedly from early in the first wave of
the pandemic [E8 and E9] to provide long term forecasts of COVID-19 spread to LNWH, and
was used during the second wave and after to provide forecasts for HH. This supported
healthcare delivery across West London in three different ways:
First, it informed the trusts’ operational response during subsequent waves of the pandemic
[E1], which led to LNWH arranging capacity in advance that benefited approximately 4,000
COVID-19 hospital patients during this period [E3]. Planning the zoning and repurposing of
wards was undertaken using a range of forecasted scenarios which included key decision or
trigger points. This was required as patient admission ranged from several per day to
approximately 1 per 10 minutes. As the pandemic progressed, models were refined and
adapted to better support general and acute (G&A) and intensive care unit (ICU) bed planning
(including surge capacity and stand-down points). This enabled the flexing of bed capacity
throughout 2020 from no COVID-19 beds up to 54 additional ICU beds and 44 acute respiratory
beds. The capacity increase also was aligned with the wider planning with local councils, such
that sufficient care home support and intermediate recovery beds were prepared for the surge
capacity scenario.
Second, it influenced the assumptions underlying the scenarios the trusts developed to help
them plan their non-COVID elective service recovery [E1]. Brunel’s forecasts provided early
warnings for new peaks which subsequently occurred, and quieter periods that occurred in
between. Both COVID-19 and elective service recovery planning affects the work pressure level
and well-being of approximately 2,600 relevant hospital staff (healthcare assistants plus
medical) and affects the care provisioning for approximately 1,600 elective inpatient and day
case admissions per week [E4]. In addition, the research helped LNWH to understand the
potential interplay of different factors in driving COVID19 admissions demand variations [E1].
Third, forecasts indicated the scale and pace at which capacity needs to be expanded and
whether it can be returned towards other health and care needs [E1]. The preparatory efforts,
informed by Brunel’s research, of both NHS trusts enabled them to facilitate more COVID-19
patients in their hospitals (629 beds occupied) during the peak of the third wave as compared to
the peak of the first wave (268 beds occupied) [E3].
The Brunel team was also requested by the Rapid Assistance In Modelling the Pandemic
(RAMP) initiative, which is coordinated by the Royal Society, to analyse the sensitivities and
limitations of the CovidSim code [REF5] using the VECMA sensitivity analysis and ensemble
simulation toolkit [REF6]. The results from Arabnejad, Groen, and others in the VECMA consortium were fed back to RAMP in a report [E7] and contributed towards a shift in SAGE away from sole reliance on the CovidSim model [E10]. The results are also published on the Science Museum web page for educational purposes to the general public [E6].
5. Sources to corroborate the impact
E 1: Testimonial and contact details of the NHS planning and transformation directors for the
LNWH NHS Trust.
E 2: Contact details of the NHS planning and transformation directors for the Hillingdon NHS
Trust.
E 3: Information on COVID-19 cases and hospitalisations in England:
01-2021.xlsx
E 4: Annual Report on LNWH Trust: https://www.england.nhs.uk/wp-
content/uploads/2019/10/London_North_West_University_Healthcare_NHS_Trust_Annual_Rep
ort_and_Accounts_2018-19.pdf
E 5: LNWH Trust Wiki page:
E 6: Science Museum blog on Virtual Pandemics:
E 7: Sensitivity Analysis and Uncertainty Quantification for the covid-sim Epidemiological Code
[26th June 2020]
E 8: Correspondence with LNWH on 13-04-2020.
E 9: Correspondence with LNWH on 24-04-2020.
E 10: Evidence of modelling presentation shift in SAGE away from reliance of single
transmission models, relative to the direct CovidSim forecasts presented by the government in
March. Attached and available at: https://www.gov.uk/government/publications/spi-m-o-covid-19-short-term-forecasts-5-august-2020
Additional contextual information
Grant funding
| Grant number | Value of grant |
|---|---|
| 800925 | £3,539,795 |
| H2020-INFRAEDI-2018 824115 | £7,079,592 |