We can uniquely examine the impacts of micro-policies on the spread of COVID-19, demonstrably informed key policy decisions in high-income settings and is also efficiently being applied to LMICs. We provide detailed guidance on which specific policy interventions, staggered at specific intervals, carried out for different durations, and relaxed back towards status quo in a staggered sequence, will most effectively mitigate COVID-19 across LMICs. We are able to explore trade-offs between epidemiology and macroeconomy for given micro-policies on industries and social behaviors.

Optima COVID-19

Background

As COVID-19 vaccines continue to be rolled-out, the combination of non-pharmaceutical interventions including physical and social distancing policies; hand hygiene and mask-wearing; testing, contact tracing, and isolation of positive cases will continue to be important in controlling the pandemic. These policies have significant impact on the lives and livelihoods of people in the countries where they are implemented, and the overall economy, so it is critical that decision-makers have the best possible information on which policies to implement and for how long, in order to achieve the greatest impact with minimal disruption to their society and economy. It is also critical that the roll back of such policies is done in a way that does not compromise overall epidemic control. The Burnet - Optima modelling team, in collaboration with the Institute for Disease Modeling, has a model that can uniquely answer these questions with regard to specific policies to provide practical guidance to LMICs.

Model

Members of the Burnet - Optima modeling team were integral in writing an agent-based COVID-19 model, Covasim, housed at the Institute for Disease Modeling. Please access the GitHub Covasim repository here. This model uses demographic data (population size, population age structure, household size distribution), setting-specific contact data (separated into household, school, work, and community contacts; from (Kerr et al. 2020) and global COVID-19 disease parameter estimates. The model is also linked with an interface developed by Microsoft that can account for economic impact. It is pre-loaded with comprehensive publicly available data (e.g., DHS) to make adaptation to different settings highly efficient. For each setting we have set-up the following scenarios:
  1. Baseline: produce epidemic projections in the absence of additional policies
  2. Baseline + control: design tailored responses to control the epidemic once it is established
    1. Estimate how effective policies would need to be to control transmission
    2. Estimate which combinations of service closures and social distancing measures could achieve this level of effectiveness, based on real-world constraints and imperfect adherence
  3. Baseline + control + release: design a multi-staged and timed approach to lifting restrictions
    1. Estimate the additional risk (probability of an outbreak and likely outbreak size) if each component of the control stage was reversed at different stages in the epidemic
    2. Identify whether strategies are available that could mitigate these risks (e.g. targeted testing)

Policies

The most unique component of this model is the capacity to take micro-policies into account, something that – as far as we are aware - none of the other available models can do. A wide range of policies can be simulated by changing the distribution of contacts (e.g., no school contacts when schools are closed, fewer community contacts when events are cancelled), or the probability of disease transmission per contact (e.g., lower risk of transmission per contact if personal protective equipment is used). Example policies include:
  • Schools: closing/opening, reducing class sizes, physical distancing in classrooms
  • Places of worship: closing/opening, social distancing
  • Bars and restaurants: closing/opening, social distancing
  • Non-essential work (nominating various specific industries) with the capacity to work from home: closing/working remotely/opening
  • Non-essential work (nominating various specific industries) without the capacity to work from home: closing/opening
  • Essential work: physical distancing, provision of personal protective equipment, distancing from non-work contacts
  • Group gatherings: ban/allow gatherings of >100, >10, >2 (or any desired number) people
  • Parks and recreational locations: closing/social distancing/opening
  • Community sports: banned/allowed
  • Professional sports: banned/without fans/allowed with fans practicing distancing/back to normal

Outputs

The model is highly dynamic and can be recalibrated to data and run daily or multiple times per week, as policy questions change or as new data become available. For settings that have an absence of data, the model is proposed to be run based on three setting archetypes: a high infection rate setting (e.g. Italy, Spain, New York), a medium infection rate setting, or a low infection rate setting. Different epidemic scales could be simulated in each LMIC to assess different control strategies. If there is desire from partners, it may be possible to also assess a formal mathematically optimal sequence of mitigation and relaxing strategy against a specific objective. Further, it may also be possible to use existing economic models to compare the potential economic impact of different micro-strategies. Lastly, it may be possible to integrate a weighted epidemic and economic objective function and optimization approach which considers both impacts. We understand the limited time to provide evidence for an emergency response and believe these features could be developed relatively quickly.