The impact of household structure on the effectiveness of shielding vulnerable populations against COVID-19 transmission: an agent-based modelling study

PROJECT BACKGROUND COVID-19 is an emerging infectious disease of global importance. As a rapidly progressing public health crisis, responsive interventions must be appropriate to the context and evolution of the epidemic. Many countries around the world have adopted widespread measures aimed at the whole population, such as national lockdowns, travel bans and extreme social distancing regulations. However, due to the high social and economic costs, these measures are not sustainable indefinitely. As these measures are relaxed or as the epidemic progresses, targeted interventions that protect the most vulnerable populations (such as the elderly and those with co-morbidities) will be required to minimise the number of severe cases and death. One of the core public health objectives of the European CDC is to "Reduce morbidity, severe disease and mortality in the population through proportionate non-medical countermeasures, with emphasis on protecting vulnerable (high-risk) groups, until effective vaccines, treatments and medicines become available." Several countries have already adopted a strategy of "shielding", which is a measure to protect extremely vulnerable people from coming into contact with coronavirus, by minimising all interaction between them and others. The definition of who should be shielded and how stringent the shielding advice is varies by setting. The majority of countries implementing targeted shielding techniques are high-income with household structures that facilitate shielding. However, it is unclear how effective this would be in settings with highly mixed populations, substantial differences between urban and rural settings, large proportions of the population living in informal settlements, and multi-generational household structures. South Africa is an example of a country with these dynamics, as well as having one of the highest GINI coefficients in the world, reflecting a deeply inequal society. While we can learn lessons from the early stages of shielding in high-income settings, understanding how these factors might influence the effectiveness of potential interventions such is critical to creating an effective, context-specific COVID-19 strategy. Mathematical modelling of infectious diseases provides a tool for policy makers and public health planners to predict the impact and cost-effectiveness of possible intervention strategies. This facilitates the effective planning and implementation of interventions, and the appropriate allocation of resources for maximum benefit. Our project will use agent-based modelling to estimate the impact of household structure on the effectiveness of shielding in South Africa. This can help inform evidence-based, context-appropriate interventions for reducing COVID-19-related hospitalisation and mortality in low- and middle-income countries. RESEARCH OBJECTIVES 1. Develop an agent-based model of COVID-19 transmission in South Africa 2. Project the number of COVID-19 cases, hospital and ICU admissions, and deaths averted under different shielding strategies (using several age and co-morbidity cut-offs) and household structures. STUDY DESIGN Study overview and setting We will develop an agent-based probabilistic simulation model coded in Python and visualised in Gephi. Compared with compartmental or population-based modelling, using an agent-based model allows significant flexibility in modelling micro-level process and individual-level behaviour. The model uses the population and age structure of South Africa. It will capture the dynamics of representative populations from different settings in South Africa incorporating the household structures and social mixing patterns (age and location based) in urban versus rural settings, and formal versus informal settlements. The model follows this population over time at an individual level through different stages of the disease. Model methodology A time varying agent-based model applying network theory will be used to capture contacts between individuals within a synthetic population. In the model, each agent is assigned to a household and an occupation, which is either work, school or "other". The agents follow a timetable assigning to them a location (household, occupation or other) for each timestep, with a parameter allowing for per-agent deviation from this timetable. On a given timestep, each place randomly assigns contacts between agents currently at this place. The probability distribution on possible contact assignments accounts for parameters such as the expected number of contacts per agent and the increased likelihood of a given agent more frequently contacting certain other agents. The agents are assigned an age, which is drawn from the South African age distribution. The disease dynamics are incorporated per agent such that each agent has a specified disease state and contacts between infectious and susceptible individuals may result in the susceptible individual moving to the exposed state. The model will be verified in several ways. One verification uses two studies of contacts in a South African population which break up the location of these contacts by place (home, school, work or other). This gives a constraint on the distribution on per-agent contacts per day per place. Another verification check uses South African Household Census data to confirm the allocation of people to households with a particular focus on household structure. An SEIR compartmental model which has already been used to assist the South African National Department of Health will be used to verify the disease dynamics component of the model. Study data The model will be informed by published and pre-print academic literature, global COVID-19 case information (specifically from the European CDC, World Health Organization and China CDC), South African population statistics from Stats SA's 2019 mid-year report, national and provincial case and testing details from the South African National Institute for Communicable Diseases and https://sacoronavirus.co.za/category/press-releases-and-notices/. As a new disease, there is much yet to be understood epidemiologically about how COVID-19 manifests and how it spreads. We will therefore conduct sensitivity analyses on key model parameters, including probability of transmission upon contact with an infectious individual and the asymptomatic proportion of infections. EXPECTED OUTCOMES & RESEARCH IMPACT The study will estimate COVID-19 cases (by age and severity), hospitalisation and ICU admissions, and mortality rates under different intervention scenarios, settings, and household structures. Results from this study will be disseminated in two ways: 1. As a publication in a peer-reviewed journal. 2. As a policy-focused report for the South African COVID-19 Modelling Consortium. Although the model is tailored to a South African context, the results can be applied to similar settings. As there is limited research on appropriate, long-term COVID-19 strategies in low- and middle-income countries, we anticipate this study will contribute to a better understanding of effective, context-specific intervention options in these settings. Where data sharing agreements allow, the model code can be made accessible to the research community. This extends the scope of research impact as it enables researchers from other countries to tailor the model to their own setting. MASHA is a member of the South African COVID-19 Modelling Consortium, a group of individuals and institutions with expertise in a range of scientific disciplines that has been convened to provide modelling support to decision makers tackling the COVID-19 epidemic in South Africa. This provides an ideal forum for presenting the study's policy-focused report and engaging decision-makers around the research results. PROJECT TEAM & ROLES Applicant/Primary researcher Rachel will be involved in all phases of the project: model development, intervention simulations, manuscript and policy brief writing. Co-researcher Jared Norman (Researcher at MASHA): Jared is a computer scientist with a background in pure and applied mathematics. His research interests are in agent-based modelling and simulation, GPU computing, and mathematical modelling of infectious diseases. He has experience in the development of user-friendly computer applications designed to allow policy makers to run simple mathematical models and navigate the output of complex models with the aid of interactive graphs. Jared's role in this project will be to provide technical support on the development of the agent-based model and simulations. Supervisor Dr Sheetal Silal (Director of MASHA): Sheetal is a mathematical modeller and statistician focusing on infectious diseases in South Africa, sub-Saharan Africa and the Asia Pacific region. Her interests are in epidemiological and economic modelling to support intervention planning and policy development. Sheetal is the founder and Director of MASHA, a senior lecturer at the University of Cape Town and an Honorary Visiting Research Fellow in Tropical Disease Modelling at the University of Oxford. Sheetal will supervise the project and be the primary contact for the policy brief submission to the South African COVID-19 Modelling Consortium.