Machine learning assisted calibration of stochastic agent-based models for pandemic outbreak analysis (preprint)

Mathematical modelling with agent-based models (ABMs) has gained popularity during the COVID-19 pandemic, but their complexity makes efficient and robust calibration to data challenging. We propose an improved method for calibrating ABMs that combines a machine-learning step with Approximate Bayesian Computation (ML-ABC). We showcase its application to Covasim - a stochastic ABM that has been timely and responsively used to model the English COVID-19 epidemic and inform policy at important junctions. We illustrate the advantage of ML-ABC application in calibrating Covasim during the first and the second COVID-19 epidemic waves of 2020 and early 2021, demonstrating that the use of an ML screening step allows us to derive faster and more efficient estimates of the posterior distribution of the Covasim optimal parameters without compromising on accuracy. This is important for generating timely responsive modelling results during an emerging epidemic.

Statistical and agent-based modelling of the transmissibility of different SARS-CoV-2 variants in England and impact of different interventions (preprint)

The English SARS-CoV-2 epidemic has been affected by the emergence of new viral variants such as B.1.177, Alpha and Delta, and changing restrictions. We used statistical models and calibration of an stochastic agent-based model Covasim to estimate B.1.177 to be 20% more transmissible than the wild type, Alpha to be 50-80% more transmissible than B.1.177 and Delta to be 65-90% more transmissible than Alpha. We used these estimates in Covasim (calibrated between September 01, 2020 and June 20, 2021), in June 2021, to explore whether planned relaxation of restrictions should proceed or be delayed. We found that due to the high transmissibility of Delta, resurgence in infections driven by the Delta variant would not be prevented, but would be strongly reduced by delaying the relaxation of restrictions by one month and with continued vaccination.

Agent-based modeling of COVID-19 outbreaks for New York state, UK and Novosibirsk region (preprint)

This paper uses Covasim, an agent-based model (ABM) of COVID-19, to evaluate and scenarios of epidemic spread in New York State (USA), the UK, and the Novosibirsk region (Russia). Epidemiological parameters such as contagiousness (virus transmission rate), initial number of infected people, and probability of being tested depend on the region's demographic and geographical features, the containment measures introduced; they are calibrated to data about COVID-19 spread in the region of interest. At the first stage of our study, epidemiological data (numbers of people tested, diagnoses, critical cases, hospitalizations, and deaths) for each of the mentioned regions were analyzed. The data were characterized in terms of seasonality, stationarity, and dependency spaces, and were extrapolated using machine learning techniques to specify unknown epidemiological parameters of the model. At the second stage, the Optuna optimizer based on the tree Parzen estimation method for objective function minimization was applied to determine the model's unknown parameters. The model was validated with the historical data of 2020. The modeled results of COVID-19 spread in New York State, the UK and the Novosibirsk region have demonstrated that if the level of testing and containment measures is preserved, the number of positive cases in New York State and the Novosibirsk region will remain the same during March of 2021, while in the UK it will reduce. Due to the features of the data for the Novosibirsk region (two datasets are characterized as stationary series with probability of 1), the forecast precision is relatively high for the number of hospitalizations, but is lower for new cases of COVID-19.

Mechanistic modeling of SARS-CoV-2 immune memory, variants, and vaccines (preprint)

Early waves of the SARS-CoV-2 pandemic were driven by importation events and subsequent policy responses. However, epidemic dynamics in 2021 are largely driven by the spread of more transmissible and/or immune-evading variants, which in turn are countered by vaccination programs. Here we describe updates to the methodology of Covasim (COVID-19 Agent-based Simulator) to account for immune trajectories over time, correlates of protection, co-circulation of different variants and the roll-out of multiple vaccines. We have extended recent work on neutralizing antibodies (NAbs) as a correlate of protection to account for protection against infection, symptomatic COVID-19, and severe disease using a joint estimation approach. We find that NAbs are strongly correlated with infection blocking and that natural infection provides stronger protection than vaccination for the same level of NAbs, though vaccines typically produce higher NAbs. We find only relatively weak correlations between NAbs and the probability of developing symptoms given a breakthrough infection, or the probability of severe disease given symptoms. A more refined understanding of breakthrough infections in individuals with natural and vaccine-derived immunity will have implications for timing of booster vaccines, the impact of emerging variants of concern on critical vaccination thresholds, and the need for ongoing non-pharmaceutical interventions.

Modelling the impact of reopening schools in early 2021 in the presence of the new SARS-CoV-2 variant and with roll-out of vaccination against COVID-19 (preprint)

Background Following the resurgence of the COVID-19 epidemic in the UK in late 2020 and the emergence of the new variant of the SARS-CoV-2 virus, B.1.1.7, a third national lockdown was imposed from January 5, 2021. Following the decline of COVID-19 cases over the remainder of January 2021, it is important to assess the conditions under which reopening schools from early March is likely to lead to resurgence of the epidemic. This study models the impact of a partial national lockdown with social distancing measures enacted in communities and workplaces under different strategies of reopening schools from March 8, 2021 and compares it to the impact of continual full national lockdown remaining until April 19, 2021. Methods We used our previously published model, Covasim, to model the emergence of B.1.1.7 over September 1, 2020 to January 31, 2021. We extended the model to incorporate the impacts of the roll-out of a two-dose vaccine against COVID-19, assuming 200,000 daily doses of the vaccine in people 75 years or older with vaccination that offers 95% reduction in disease acquisition and 10% reduction of transmission blocking. We used the model, calibrated until January 25, 2021, to simulate the impact of a full national lockdown (FNL) with schools closed until April 19, 2021 versus four different partial national lockdown (PNL) scenarios with different elements of schooling open: 1) staggered PNL with primary schools and exam-entry years (years 11 and 13) returning on March 8, 2021 and the rest of the schools years on March 15, 2020; 2) full-return PNL with both primary and secondary schools returning on March 8, 2021; 3) primary-only PNL with primary schools and exam critical years (Y11 and Y13) going back only on March 8, 2021 with the rest of the secondary schools back on April 19, 2021 and 4) part-Rota PNL with both primary and secondary schools returning on March 8, 2021 with primary schools remaining open continuously but secondary schools on a two-weekly rota-system with years alternating between a fortnight of face-to-face and remote learning until April 19, 2021. Across all scenarios, we projected the number of new daily cases, cumulative deaths and effective reproduction number R until April 30, 2020. Results Our calibration across different scenarios is consistent with the new variant B.1.1.7 being around 60% more transmissible. Strict social distancing measures, i.e. national lockdowns, are required to contain the spread of the virus and control the hospitalisations and deaths during January and February 2021. The national lockdown will reduce the number of cases by early March levels similar to those seen in October with R also falling and remaining below 1 during the lockdown. Infections start to increase when schools open but if other parts of society remain closed this resurgence is not sufficient to bring R above 1. Reopening primary schools and exam critical years only or having primary schools open continuously with secondary schools on rotas will lead to lower increases in cases and R than if all schools open. Under the current vaccination assumptions and across the set of scenarios considered, R would increase above 1 if society reopens simultaneously, simulated here from April 19, 2021.Findings Our findings suggest that stringent measures are necessary to mitigate the increase in cases and bring R below 1 over January and February 2021. It is plausible that a PNL with schools partially open from March 8, 2021 and the rest of the society remaining closed until April 19, 2021 may keep R below 1, with some increase evident in infections compared to continual FNL until April 19, 2021. Reopening society in mid-April, with the vaccination strategy we model, could push R above 1 and induce a surge in infections, but the effect of vaccination may be able to control this in future depending on the transmission blocking properties of the vaccines.

Controlling COVID-19 via test-trace-quarantine (preprint)

Initial COVID-19 containment in the United States focused on limiting mobility, including school and workplace closures. However, these interventions have had enormous societal and economic costs. Here we demonstrate the feasibility of an alternative control strategy, test-trace-quarantine: routine testing of primarily symptomatic individuals, tracing and testing their known contacts, and placing their contacts in quarantine. We performed this analysis using Covasim, an open-source agent-based model, which was calibrated to detailed demographic, mobility, and epidemiological data for the Seattle region from January through June 2020. With current levels of mask use and schools remaining closed, we found that high but achievable levels of testing and tracing are sufficient to maintain epidemic control even under a return to full workplace and community mobility and with low vaccine coverage. The easing of mobility restrictions in June 2020 and subsequent scale-up of testing and tracing programs through September provided real-world validation of our predictions. Although we show that test-trace-quarantine can control the epidemic in both theory and practice, its success is contingent on high testing and tracing rates, high quarantine compliance, relatively short testing and tracing delays, and moderate to high mask use. Thus, in order for test-trace-quarantine to control transmission with a return to high mobility, strong performance in all aspects of the program is required.

Modelling the impact of reopening schools in early 2021 in the presence of the new SARS-CoV-2 variant and with roll-out of vaccination against COVID-19 (preprint)

Background Following the resurgence of the COVID-19 epidemic in the UK in late 2020 and the emergence of the new variant of the SARS-CoV-2 virus, B.1.1.7, a third national lockdown was imposed from January 5, 2021. Following the decline of COVID-19 cases over the remainder of January 2021, it is important to assess the conditions under which reopening schools from early March is likely to lead to resurgence of the epidemic. This study models the impact of a partial national lockdown with social distancing measures enacted in communities and workplaces under different strategies of reopening schools from March 8, 2021 and compares it to the impact of continual full national lockdown remaining until April 19, 2021. Methods We used our previously published model, Covasim, to model the emergence of B.1.1.7 over September 1, 2020 to January 31, 2021. We extended the model to incorporate the impacts of the roll-out of a two-dose vaccine against COVID-19, assuming 200,000 daily doses of the vaccine in people 75 years or older with vaccination that offers 95% reduction in disease acquisition and 10% reduction of transmission blocking. We used the model, calibrated until January 25, 2021, to simulate the impact of a full national lockdown (FNL) with schools closed until April 19, 2021 versus four different partial national lockdown (PNL) scenarios with different elements of schooling open: 1) staggered PNL with primary schools and exam-entry years (years 11 and 13) returning on March 8, 2021 and the rest of the schools years on March 15, 2020; 2) full-return PNL with both primary and secondary schools returning on March 8, 2021; 3) primary-only PNL with primary schools and exam critical years (Y11 and Y13) going back only on March 8, 2021 with the rest of the secondary schools back on April 19, 2021 and 4) part-Rota PNL with both primary and secondary schools returning on March 8, 2021 with primary schools remaining open continuously but secondary schools on a two-weekly rota-system with years alternating between a fortnight of face-to-face and remote learning until April 19, 2021. Across all scenarios, we projected the number of new daily cases, cumulative deaths and effective reproduction number R until April 30, 2020. Results Our calibration across different scenarios is consistent with the new variant B.1.1.7 being around 60% more transmissible. Strict social distancing measures, i.e. national lockdowns, are required to contain the spread of the virus and control the hospitalisations and deaths during January and February 2021. The national lockdown will reduce the number of cases by early March levels similar to those seen in October with R also falling and remaining below 1 during the lockdown. Infections start to increase when schools open but if other parts of society remain closed this resurgence is not sufficient to bring R above 1. Reopening primary schools and exam critical years only or having primary schools open continuously with secondary schools on rotas will lead to lower increases in cases and R than if all schools open. Under the current vaccination assumptions and across the set of scenarios considered, R would increase above 1 if society reopens simultaneously, simulated here from April 19, 2021. Findings Our findings suggest that stringent measures are necessary to mitigate the increase in cases and bring R below 1 over January and February 2021. It is plausible that a PNL with schools partially open from March 8, 2021 and the rest of the society remaining closed until April 19, 2021 may keep R below 1, with some increase evident in infections compared to continual FNL until April 19, 2021. Reopening society in mid-April, with the vaccination strategy we model, could push R above 1 and induce a surge in infections, but the effect of vaccination may be able to control this in future depending on the transmission blocking properties of the vaccines.

Preventing a cluster from becoming a new wave in settings with zero community COVID-19 cases (preprint)

In settings with zero community transmission, any new SARS-CoV-2 outbreaks are likely to be the result of random incursions. The level of restrictions in place at the time of the incursion is likely to considerably affect possible outbreak trajectories. We used an agent-based model to investigate the relationship between ongoing restrictions and behavioural factors, and the probability of an incursion causing an outbreak and the resulting growth rate. We applied our model to the state of Victoria, Australia, which has reached zero community transmission as of November 2020. We found that a future incursion has a 45% probability of causing an outbreak (defined as a 7-day average of >5 new cases per day within 60 days) if no restrictions were in place, decreasing to 23% with a mandatory masks policy, density restrictions on venues such as restaurants, and if employees worked from home where possible. A drop in community symptomatic testing rates was associated with up to a 10-percentage point increase in outbreak probability, highlighting the importance of maintaining high testing rates as part of a suppression strategy. Because the chance of an incursion occurring is closely related to border controls, outbreak risk management strategies require an integrated approaching spanning border controls, ongoing restrictions, and plans for response. Each individual restriction or control strategy reduces the risk of an outbreak. They can be traded off against each other, but if too many are removed there is a danger of accumulating an unsafe level of risk. The outbreak probabilities estimated in this study are of particular relevance in assessing the downstream risks associated with increased international travel.

Lessons learned from Vietnam’s COVID-19 response: the role of adaptive behavior change and testing in epidemic control (preprint)

Background Vietnam has emerged as one of the world’s leading success stories in responding to COVID-19. After prolonged zero-low transmission, a summer outbreak of unknown source at Da Nang caused the country’s first COVID-19 deaths, but was quickly suppressed. Vietnam recently reopened its borders to international travelers. Understanding the attendant risks and how to minimize them is crucial as Vietnam moves into this new phase. Methods We create an agent-based model of COVID-19 in Vietnam, using regional testing data and a detailed linelist of the 1,014 COVID-19 cases, including 35 deaths, identified across Vietnam. We investigate the Da Nang outbreak, and quantify the risk of another outbreak under different assumptions about behavioral/policy responses and ongoing testing. Results The Da Nang outbreak, although rapidly contained once detected, nevertheless caused significant community transmission before it was detected; higher symptomatic testing could have mitigated this. If testing levels do not increase, the adoption of past policies in response to newly-detected cases may reduce the size of potential outbreaks but will not prevent them. Compared to a baseline symptomatic testing rate of 10%, we estimate half as many infections under a 20% testing rate, and a quarter as many with 40-50% testing rates, over the four months following border reopenings. Conclusions Vietnam’s success in controlling COVID-19 is largely attributable to its rapid response to detected outbreaks, but the speed of response could be improved even further with higher levels of symptomatic testing.

Determining the optimal strategy for reopening schools, work and society in the UK: balancing earlier opening and the impact of test and trace strategies with the risk of occurrence of a secondary COVID-19 pandemic wave (preprint)

Background In order to slow down the spread of SARS-CoV-2, the virus causing the COVID-19 pandemic, the UK government has imposed strict physical distancing (lockdown) measures including school 'dismissals' since 23 March 2020. As evidence is emerging that these measures may have slowed the spread of the pandemic, it is important to assess the impact of any changes in strategy, including scenarios for school reopening and broader relaxation of social distancing. This work uses an individual-based model to predict the impact of a suite of possible strategies to reopen schools in the UK, including that currently proposed by the UK government. Methods We use Covasim, a stochastic agent-based model for transmission of COVID-19, calibrated to the UK epidemic. The model describes individuals' contact networks stratified as household, school, work and community layers, and uses demographic and epidemiological data from the UK. We simulate a range of different school reopening strategies with a society-wide relaxation of lockdown measures and in the presence of different non-pharmaceutical interventions, to estimate the number of new infections, cumulative cases and deaths, as well as the effective reproduction number with different strategies. To account for uncertainties within the stochastic simulation, we also simulated different levels of infectiousness of children and young adults under 20 years old compared to older ages. Findings We found that with increased levels of testing of people (between 25% and 72% of symptomatic people tested at some point during an active COVID-19 infection depending on scenarios) and effective contact-tracing and isolation for infected individuals, an epidemic rebound may be prevented across all reopening scenarios, with the effective reproduction number (R) remaining below one and the cumulative number of new infections and deaths significantly lower than they would be if testing did not increase. If UK schools reopen in phases from June 2020, prevention of a second wave would require testing 51% of symptomatic infections, tracing of 40% of their contacts, and isolation of symptomatic and diagnosed cases. However, without such measures, reopening of schools together with gradual relaxing of the lockdown measures are likely to induce a secondary pandemic wave, as are other scenarios for reopening. When infectiousness of <20 year olds was varied from 100% to 50% of that of older ages, our findings remained unchanged. Interpretation To prevent a secondary COVID-19 wave, relaxation of social distancing including reopening schools in the UK must be implemented alongside an active large-scale population-wide testing of symptomatic individuals and effective tracing of their contacts, followed by isolation of symptomatic and diagnosed individuals. Such combined measures have a greater likelihood of controlling the transmission of SARS-CoV-2 and preventing a large number of COVID-19 deaths than reopening schools and society with the current level of implementation of testing and isolation of infected individuals.