key: cord-0907314-t16feue3 authors: Groen, B. M.; Turner, P.; Lacey, P. title: The expediency of local modelling to aid national responses to SARS-CoV-2. date: 2020-05-29 journal: nan DOI: 10.1101/2020.05.27.20107656 sha: ac1aa2995f7e20586b13b1f433709a27eff01f0f doc_id: 907314 cord_uid: t16feue3 Background: With the SARS-CoV-2 pandemic gripping most of the globe, healthcare and economic recovery strategies are being explored currently as a matter of urgency. Methods: We adopted the epidemiological model which informs United Kingdom government's national policy on the SARS-CoV-2 pandemic and used it in a local context and show how projections in terms of presentations of symptomatic patients in a variety of settings differ depending on local demographics. Setting: England, United Kingdom, population level modelled 3.2m Results: We clearly demonstrate that there is significant difference in the way the modelling behaves in terms of potential subsequent waves for levels of demand on the health and care systems. This provides evidence of differing timeline of peak demands and clearly shows varying levels of demand throughout the four modelled sub-geographies. Conclusions: The study results clearly outline the expediency of national modelling being adopted to reflect local health and care systems. We urge readers to ensure that any national policy is appropriately adopted to suit local health and care systems. In addition, we share our methodology to ensure other professionals could replicate this study elsewhere. Background National governments are forced to take urgent action with national policy restricting social interactions and businesses to temporarily close. In order to fully understand how the limited health and care resources could be targeted in response to this pandemic, we need to create a thorough understanding of the impact of national policy on local health and care systems. The guiding principle and understanding here is that health and care services are provided locally not nationally, not in aggregate but to individuals. Therefore, we conducted a study using existing models which are used to inform national policy and adapted these to consider local nuances to see whether such projective model would behave differently; showing how projections in terms of presentations of symptomatic patients in a variety of settings differ depending on local demographics. 2 Background 1 National governments are forced to take urgent action using national policies to restrict social 2 interactions and temporary business closures which impact significantly on numerous factors of our 3 lives. In order to fully understand how limited health and care resources could be utilised and 4 maximised in response to this pandemic, we need to create a thorough understanding of the impact 5 of national policy on local health and care systems. Our guiding principle and understanding here is 6 that health and care services are provided locally (effectively at Integrated Care Partnership -ICP 7 level) not regionally or nationally, indeed, not in aggregate but to individuals. Therefore, we conducted 8 a study using well-established model parameters which are used to inform national policy and adapted 9 these to consider local nuances to see whether any projected model outputs would behave differently 10 to national projections. We show how projections in terms of presentations of symptomatic patients 11 in a hospital setting significantly vary between local communities. Indeed, when aggregated up to 12 regional or national levels, such local idiosyncrasies fade which may have profound consequences in 13 efforts to coordinating local health and care services in response to SARS-CoV-19. 14 Model design 17 A System Dynamic Model (SDM) approach is selected and a Susceptible, Exposed, Infectious, 18 Recovered, (SEIR) stock and flow model designed by the SDM software manufacturer 4 is adapted for 19 the base model. The model is run over a one-year period starting from 1 st January 2020 and calculates 20 one transaction per day using the Euler integration method. SDM offers communities a tool with which 21 to understand their systems and become ready to influence and engage with real-world actions 22 (Minyard et al., 2018) whilst avoiding discrete operational level interference. The purpose of the SDM 23 allows Integrated Care System (ICS) communities to simulate scenarios for non-pharmaceutical 24 interventions to SARS-CoV-2 and examine care provision away from a national perspective. Porter 25 and Oleson discuss limitations in SEIR arising from exponential distribution of latent and infectious 26 times (Porter & Oleson, 2013) The SDM is calibrated with data describing the population, virus characteristics and actual 4 outcomes. Population data are extracted from the Office for National Statistics (ONS) census 5 projections by age and Clinical Commissioning Group (CCG) and applied to the ICS or population being 6 modelled (see Localisation sector for further work). The virus characteristics are taken from the 7 Imperial College COVID-19 Response Team (ICCRT) simulation, which informs United Kingdom 8 government's national policy on the SARS-CoV-2 pandemic (Ferguson et al., 2020) ; namely the fatality 9 and virus transmission rates by age group. Actual outcomes, for daily hospital deaths and admissions 10 are taken from ICS situation reports which are consistent with NHS England reporting and hospital 11 length of stay data are extracted from each acute hospital trust. Generic processed data is available 12 by consulting (Groen & Turner, 2020 Progression of the disease: 1. Recovery -possible at each stage of the model, but different proportions at asymptomatic and symptomatic stages and also for each risk group. 2. Progression -the rate is determined by the time in days between infection and symptoms appearing, and then to needing hospital admission or recovering, different %'s are applied for each risk group and at each stage. 3. Death -is possible at infected stage either in hospital or community (currently national modelling only estimates deaths from hospital, and within hospital from CCU beds). All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 29, 2020. . https://doi.org/10.1101/2020.05.27.20107656 doi: medRxiv preprint Adaptation and Localisation 1 Using the ICCRT simulation, the SDM categorised the initial susceptible population into age 2 groupings. To reflect local characteristics, this was augmented to include a risk profile (low, moderate, 3 and high) using population-based health profiles taken from a separate proprietary model (the Cohort 4 model). The cohort model uses data extracted from ONS and the Kent Integrated Dataset to create 5 prevalence and incidence for morbidities (including expected multi-morbidities) adjusted for each 6 English CCG. Models were calibrated for each ICS in the North East and North Cumbria region of 7 England, based on its constituent CCGs, and was seeded for initial infections 30 days prior to the first 8 cluster of deaths on sequential days and validated so that modelled outcomes for ICS hospital bed 9 occupation and deaths fitted actual situation report results. An example ICS results are shown in figure 10 2 and 3. 11 (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 29, 2020. . https://doi.org/10.1101/2020.05.27.20107656 doi: medRxiv preprint Scenarios 1 We set four scenarios to simulate the demand on healthcare service provision following the 2 initial intervention to reduce potential virus transmitting contacts introduced by the UK government 3 during March 2020. Baseline contacts pre-intervention (before March 2020) were set at a nominal 4 value of 100 and adjusted for likely reductions from government interventions for the remainder of 5 2020. The scenarios tested. The adjusted contact value and timings for these scenarios are 6 summarised in table 1 below. 7 Description Scenario 1 The initial impact from the intervention is extended but weakens following initial success. Later in the year the impact from the intervention increases as a result of effective test, track, and trace. Gradual, medium to long term relaxation of social distancing from May 10th results in a reduction in the effectiveness of social distancing. Cyclical relax/renew -May 10th relaxation with a reduction in the effectiveness of social distancing following by subsequent lock down and release. Extended social distancing but weakening of the initial lock down followed by effective implementation of track and trace. The scenarios were run for four local communities in the North East of England and North 12 Cumbria. All calibration data were assumed the same except for population health needs and the 13 seeding date required for timing in order to fit the model outcomes to actual. Modelled Outcomes 14 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 29, 2020. . https://doi.org/10.1101/2020.05.27.20107656 doi: medRxiv preprint from the scenarios were analysed for hospital deaths, acute hospital bed occupancy split between bed 1 capacity requiring ventilation, continuous positive airway pressure (oxygen plus) and oxygen. 2 Epidemiological progression was also analysed using: 3 • the effective reproduction number -Re (the proportion of overall population remaining 4 susceptible to the virus). Re = Ro x (population susceptible / total population) 5 • the effective reproduction number over time -Rt (representing the average new infections 6 arising from active infections at timet). 7 Epidemiology 8 The differences in susceptible population resulting from the pre-intervention viral spread can 9 be seen in the results in figure 4 . The efficacy of the changing contact rate arising from the scenarios 10 creates dynamics in viral progression, for example West (i.e. North Cumbria) recorded earlier cases in 11 larger quantities than the other areas within the ICS and the effect can be seen in Re lowering at a 12 more rapid rate in the period to April 2020. All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 29, 2020. Results above are split by locality in figure 6 modelling demand for oxygen beds. This 7 illustrates overall demand for this type of bed, however, we have modelled this for each type of SARS-8 CoV-19 demand a more detailed analysis and anonymised data can be accessed at (Groen & Turner, 9 2020 ). Generally speaking the n beds is based on the , which accounts for the higher number of total 10 demand for such bed types. However, it is important to note that patients admitted to this type of 11 beds tend to have a shorter length of stay and are clinically less complicated to manage operationally. 12 13 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 29, 2020. The model is indicating a higher first peak in terms of hospital deaths, owing to the fact that 5 the cohort approach considers the remaining susceptibility amongst the uninfected population and 6 stratifies this by category of cohort. 7 Discussion 8 With regards to contact rate and seeding of the model -we have subjected our approach to 9 numerous tests to demonstrate that the overall outcomes from the pandemic, for example in terms 10 of deaths, are highly sensitive to the timing of initial seeding and the subsequent lock-down. We 11 applied the national model on a single data file on the 23rd March 2020. In areas of high initial seeding 12 of the virus the additional days of spread may result in differences in the overall number of deaths 13 expected and in the subsequent size of any second wave of the pandemic. This will rely on different 14 levels of the remaining susceptible populations after the observed initial peak. We certainly believe 15 that this strengthens the case for scenario planning of subsequent easing or resumption of any lock-16 down measures on the basis our locally focused approach to modelling this pandemic. 17 The model is sensitive to two major variables outside of the virus characteristic, these are 18 seeding volume/date and the contact reduction. Model seeding assumes a certain number of 19 infections entering the region to create the virus uptake prior to the first peak, however it does not 20 currently assume any new cases coming from out of region after that time. This seems an unlikely 21 scenario and future version of the model will address this issue as more evidence on travel following 22 easing of government policy is made we aim to refine this using intelligence and conceptual 23 approaches such as outlined in (China CDC, 2020 ). This will be important as the current 24 epidemiological element of the model is one of self-perpetuating transmission -the current infected 25 population are the only ones who will contribute to new infections. Social distancing contacts also 26 depend on a less aggregated view. Local characteristics for potentially virus transmitting contacts can 1 be estimated through data including urban / rural or population density measures (Leung et al., 2020; 2 Li et al., 2020), household density, vulnerability indices and movement/transport (Kraemer et al., 3 2020) see (Prem et al., 2020) for how these factors played a role in China's approach to social isolation. 4 This would allow greater understanding of influences on clinical demand through more sophisticated 5 scenario building. 6 As per well-established literature, the model concurs on recent peer-reviewed publications 7 and popular media outlet coverage which draw attention to a 'second wave' scenario, see (World 8 Health Organization, 2020) and (Xu & Li, 2020) for example. Indeed, it is within that context that we 9 stress the importance of locally defined modelling approach (effectively; bottom up) which reflects 10 local population needs, which, when aggregated will comprise a more insightful and nuanced 11 approach to inform national approaches to this global challenge. We call on researchers to adopt our 12 approach within their own local context to ensure health and care demands are met locally within the 13 inevitable constraints that comes with national policy. 14 15 Contact too sensitive -developed into multidimensional transmission by place, movement 17 vulnerability, social/domestic contact which will be applied to the age-based population health needs. 18 Questions over the small number needed for seeding compared to local estimates (may mean 19 infection rate is higher as per other papers). Model fitted to recorded hospital admissions and deaths 20 (although COVID deaths outside of hospital are modelled. 21 • Contact rate and seeding -highly sensitive independent factors 22 • Deaths peak in first wave -more high risk / vulnerable (from cohort model) 23 • Beds have generally higher demand in second (subsequent waves, depending on relaxation in 24 scenario) 25 • Time of initial infections important in epidemiology. 26 • The capacity numbers used in the reports remain high as they are in use nationally at this level 27 but we will look to work with providers to add more 'accurate' numbers in the coming week, 28 alongside those currently in use. 29 30 All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 29, 2020. . https://doi.org/10.1101/2020.05.27.20107656 doi: medRxiv preprint The Novel Coronavirus Pneumonia Emergency Response Epidemiology 2 Team. The epidemiological characteristics of an outbreak of Impact 7 of non-pharmaceutical interventions (NPIs) to reduce COVID-19 mortality and healthcare 8 demand SARS-CoV-19 North East and North Cumbria Data -Harvard 10 The effect of human mobility and control 12 measures on the COVID-19 epidemic in China First-wave COVID-19 transmissibility and severity 14 in China outside Hubei after control measures, and second-wave scenario planning: a modelling 15 impact assessment Early 19 Transmission Dynamics in Wuhan, China, of Novel Coronavirus-Infected Pneumonia A modified SEIR model to predict the COVID-19 outbreak in Spain: 22 simulating control scenarios and multi-scale epidemics The Effectiveness of Social Distancing in Mitigating COVID-19 Spread: a 25 modelling analysis Community and 28 programmatic factors influencing effective use of system dynamic models A Path-Specific SEIR Model for use with General Latent and 31 Infectious Time Distributions The effect of control strategies to reduce social mixing 34 on outcomes of the COVID-19 epidemic in Wuhan, China: a modelling study Rolling updates on coronavirus disease (COVID-19): WHO 37 characterizes COVID-19 as a pandemic Beware of the second wave of COVID-19