In this guest blog Dr Mark Hanly, Oisin Fitzgerald and Dr Tim Churches from the Centre for Big Data Research in Health at UNSW Sydney discuss the COVID-19 Open-source Infection Dynamics (COVOID) project, an R package they developed to model COVID-19 and other infectious diseases using deterministic compartmental models (DCMs).
At the outset of the COVID-19 pandemic, mathematical models were used to great effect by numerous research teams around the world to quantify the severe consequences of an unmitigated spread of the SARS-CoV-2 virus. Notably, such modelling work from the Imperial College London was instrumental in steering the UK government towards prudent lockdown measures following a brief flirtation with a herd immunity strategy. 
As stringent suppression measures bring COVID-19 transmission under control, the policy-focus naturally turns to lockdown exit strategies. In the absence of a safe and effective vaccine, governments must steer a course between the pressing social and economic motivations to reopen society and the risk of a second wave compelling further undesirable restrictions, as already witnessed in Spain, Japan and most recently in Melbourne, Australia. As at the start of the pandemic, sophisticated modelling will be needed to navigate this difficult policy terrain.
To meet this modelling challenge, our team of health data scientists at the University of New South Wales has developed the COVOID (COVID-19 Opensource Infection Dynamics) package. COVOID implements a range of extended SIR (Susceptible – Infectious – Recovered) epidemic models that allow users to explore infection spread while crucially also incorporating flexible, time-varying policy interventions. This feature allows users to first calibrate their models to existing incidence data based on the actual interventions applied in their country over the past few months. Users can then specify and compare projected outcomes under a range of potential future policy options.
The COVOID package implements two distinct classes of intervention: measures that target the number of daily social contacts between individuals, such as social distancing and quarantining; and measures that target the probability of transmission, such as hand-washing or wearing a mask. These interventions can be applied in isolation or simultaneously and can be flexibly scaled up and down over time to emulate dynamic raising and lowering of restrictions. In addition, interventions can optionally be targeted at specific subgroups, such as contacts taking place at schools or at workplaces, to emulate school closures or working from home.
How our societies respond to the pandemic has profound health, economic and social consequences and accordingly the models and assumptions that guide these critical decisions should be open, transparent and debatable. With this in mind, COVOID has been expressly designed with open source and reproducible workflows in mind. The package is implemented in the free and widely used R data science platform and includes an interactive, user-friendly app that supports non-specialists to quickly engage with the functionality without having to download or configure software. To encourage reproducibility in this interactive mode, the app includes a report canvas feature that allows the user to save interesting model results as they work with the tool. It is then possible to comment on these saved results and, with the click of a button, compile them into a simple report that documents the model outcomes as well as the assumed parameters underpinning the findings.
We have included a variety of features to support international users to apply COVOID in their country or region. Several of the SIR models implemented in the package are age-structured models, meaning that they can account for age distributions and patterns of social contact between age groups that are key to infection transmission and vary widely between high-, middle- and low-income countries. The package is also shipped with a variety of open data sources included, namely population totals, age distributions and estimated social contact patterns for over 150 countries, so that models can be easily tailored to a particular setting.
In June, COVOID was selected as joint winner of CovidR, a contest of R contributions to the COVID-19 pandemic featured as part of the European R Users Meeting (eRUM) 2020. You can view a recording of our eRUM202 presentation describing the COVOID package and interface here. R users can download the COVOID package development version from GitHub, and are welcome to contribute issues, comments, or feature requests via the GitHub Issues page. Instructions, examples and detailed package vignettes are available here.
As the coronavirus pandemic continues to evolve so too will the key policy questions facing policy-makers. From serological testing to vaccine deployment strategies, we aim to adapt to unfolding events and continue to develop tools to support policy decision-making throughout the mid- and end-stage pandemic phase.


Please Sign in (or Register) to view further.