key: cord-0483622-tidqvkvt
authors: Marquioni, Vitor M.; Aguiar, Marcus A.M. de
title: Quantifying the effects of quarantine using an IBM SEIR model on scalefree networks
date: 2020-05-28
journal: nan
DOI: nan
sha: f9184af395f11295d1979b8c9bdba7e8debfc168
doc_id: 483622
cord_uid: tidqvkvt

The COVID-19 pandemic led several countries to resort to social distancing, the only known way to slow down the spread of the virus and keep the health system under control. Here we use an individual based model (IBM) to study how the duration, start date and intensity of quarantine affect the height and position of the peak of the infection curve. We show that stochastic effects, inherent to the model dynamics, lead to variable outcomes for the same set of parameters, making it crucial to compute the probability of each result. To simplify the analysis we divide the outcomes in only two categories, that we call {best and worst scenarios. Although long and intense quarantine is the best way to end the epidemic, it is very hard to implement in practice. Here we show that relatively short and intense quarantine periods can also be very effective in flattening the infection curve and even killing the virus, but the likelihood of such outcomes are low. Long quarantines of relatively low intensity, on the other hand, can delay the infection peak and reduce its size considerably with more than 50% probability, being a more effective policy than complete lockdown for short periods.

(IBM) and studied how different types of quarantine change the infection dynamics. Individuals are modeled as nodes of a scale-free (Barabási-Albert) network [20] that can only infect their connected neighbors. Because the dynamics is stochastic, independent simulations with the same set of parameters can lead to quite different outcomes. Here we group the outcomes in only two categories, that we call best and worst scenarios. Stochasticity, a reality of the Sars-Cov-2 infection, is not captured by the mean field approximation of the SEIR model, [21] where outcomes depend deterministically on the model parameters.

We find three types of quarantine that can be effective against the epidemic: (i) relatively long (10 weeks) and intense (more than 80% isolation); (ii) short (8 weeks) and of intermediate intensity (around 70% isolation) and; (iii) long (12 weeks or more) with relatively low intensity (40% isolation). The first type, which completely ends the epidemic, is clearly the best but also the most difficult to achieve. The second type is feasible, but we find that most of the times (in most of the simulations) they result in worst case scenarios. The third type emerges as the most practical and easy to apply. It is not so effective as the previous types, but does decrease the infection peak by half. Also, it falls into the best case scenario more than 50% of the times and even in the worst scenarios the infection peak decreases.

We model the spread of the virus using an extension of the SEIR model (susceptible, exposed, infected and recovered (or removed)). Exposed individuals simulate the incubation period of the disease, when infected subjects cannot yet pass on the virus. The mean field version of SEIR model is described by the equationṡ

where N is the population size, β is the infection rate, σ the rate at which exposed become infected and γ the recovery rate. The basic reproductive number, R 0 = β/γ, gives the number of secondary infections generated by the first infectious individual over the full course of the epidemic in a fully susceptible population. (c) one neighbor did get the virus and become exposed (orange); (d) neighbors of first infected individual might still get the virus, but the orange node is still in incubation time; (e) another node gets the virus from the first infected becoming exposed and the old exposed becomes infected; (f) more nodes might get the virus (yellow) and; (g) some do become exposed while the first infected becomes recovered.

In order to take into account heterogeneity in the number of contacts we use instead an individual based model where the population is represented by a Barabási-Albert network [20, 22, 23] with N of nodes and an average degree D. As in a deterministic SEIR model, the nodes can be classified as susceptible, exposed, infected and recovered. Susceptible individuals can become exposed if connected to an infected one; exposed individual i becomes infected after a period t i of virus incubation; infected individuals can recover, and once recovered it is considered immune and therefore cannot be infected again.

At the beginning of the simulation, only one node, chosen at random, is infected while all the remaining population is susceptible. Every susceptible node connected to the infected individual becomes exposed with the transmission probability p whereas the infected node might itself recover with probability r. The probability p can be calculated

where τ symp is the average time duration of symptoms. We assume that the symptoms last for a time τ distributed according to an exponential distribution F(τ ) = λe −λτ . Once a node is exposed, it stays exposed for an incubation time t i , chosen according to a given distribution P(t i ). [24] After this period it becomes infected and is able to infect other nodes. It follows that r ≈ λ = 1/τ symp , for small λ. For P(t i ) we have used a Γ(α, β) distribution, as in [25] . beginning t s = 20, 30, and 40 days after the first infected node appears (at the beginning of the simulation). The results were divided in two different scenarios, the best and the worst cases. For each set of parameters, the best scenario consists of simulations where the infection peak is lower than the average peak of the full set of simulations, whereas the worst scenario contains the set with higher than average peaks. This approach is important because in many cases the epidemic response to the quarantine is not satisfactory, and 5 this might be solely due to stochastic effects, a common feature of real systems. As an example, Fig. 2 shows the evolution curves of infected plus exposed individuals for all 25 replicas for Q = 0.9 and t d = 10 weeks. Since independent populations, represented by different Barabási-Albert networks generated with the same specifications, under the same quarantine parameters might respond drastically different to quarantine, we also need to know the probability of each outcome. This information is complemented by Fig. 4 , that shows how peak center changes with 7 quarantine parameters, and Fig. 5 , displaying the proportion of recovered individuals at the end of the epidemic. The purple ellipse in Fig. 3 marks the parameter region where quarantine is very intense and lasts for more than 8 weeks, an ideal situation that works around 90% of the times but is very hard to enforce in practice. In this case the epidemic stops quickly (blue areas in Fig. 4 ) and less than 10% of the population is infected (green areas in Fig. 5 ).

The red ellipse in Fig. 3 shows a transition zone where the best scenario corresponds to substantial curve flattening. The center of the red ellipse is at Q ≈ 0.5 for t s = 20 but shifts to Q ≈ 0.9 for t s = 40, showing the importance of starting quarantine early. For all values of t s the red ellipse is centered at t d ≈ 6 weeks, which is a relatively short duration. Peak center, however, is not delayed in the best case scenarios. Importantly, best case scenarios are very unlikely in this region, occurring with probability around 20%.

Finally, the region surrounded by the green ellipse in Fig. 3 corresponds to long but moderate intensity quarantines. For the three values of t s considered peak height was reduced by about 50% in the best case scenarios, which happens about 50% of the times. Peak center was not significantly delayed in the best scenarios, but was pushed forward in the worst scenarios, where peak height was reduced to about 70% with respect to non-quarantine height. Interestingly, in both scenarios about 70% of the population was infected at the end of the simulation, showing that herd immunity was achieved (corresponding to the pink areas in Fig 5) .

Quarantine can also be implemented in the mean field model, Eqs. (1) . [27] This is accomplished by integrating the dynamical equations with the infection rate β 0 for t ∈ [0, t s ], with the reduced value β Q = (1 − Q)β 0 during quarantine period t s < t < ts + t d and again with β 0 for t > t s + t d . Fig. 6 In order to distinguish between different outcomes we have divided them into two groups with the best and worst results based on the height of the infection peak (below or above the average height, respectively).

We have further divided the results into four qualitative classes delimited by the three ellipses in Fig. 3 plus the rest of the diagram. Besides the obvious region indicated by the purple ellipse where quarantine is very intense and long, we found that short but not so intense quarantine (red ellipse) does not work, since the probability of an outcome in the best scenario is very low. Instead, long but average intensity quarantine is both likely to work and flattens the infection curve by around 50%, being the best alternative given the current assumptions. Indeed, the infection peak is considerably delayed in the region of the green ellipse when it falls into the worst scenario, confirming it as the best bet for preventing the health system breakdown (Fig. 4 ) . The proportion of the population that had contact with the virus at the end of the epidemic (number of recovered individuals, Fig. 5 ) leads to more than 60% of the population, very close to achieving herd immunity. Comparing to the other regions, this seems to be the best option to control the epidemics under the model assumptions. We note, however, that the model does not account for deaths. If achieving herd immunity implies high mortality, the best option would be long and intense quarantine (purple ellipses in Fig. 3 ), the only way to avoid large number of infections and, therefore.

high mortality.

We found that differences between mean field and stochastic models are very significant with respect to the effects of quarantine. In many cases as the former cannot control the epidemic, as the infection peak grows again once the quarantine period is over, whereas the latter can end the epidemic in the best case scenarios.

We recall that we used uniform decrease in infection rate as a proxy for quarantine. This is a simplified approach and other methods could be implemented to verify the robustness of the results. Also, different network topologies might affect the spread of the epidemics.

Random uniform (Erdos-Renyi) [22] networks should produce results similar to mean field simulations, but small-world [22, 29] or other topologies could speed up or slow down the spread dynamics.

Our model is particularly suited to study spread between connected cities, that can be represented by modules of a larger network. We have also kept information about the virus DNA and its mutations, allowing us to reconstruct the phylogeny and classify its strains as it propagates. These results will be published in a forthcoming article.

the São Paulo Research Foundation (FAPESP), grants 2019/13341-7 (VMM), 2019/20271 and 2016/01343-7 (MAMA), and by Conselho Nacional de Desenvolvimento Científico e Tecnológico

A novel coronavirus from patients with pneumonia in china

World Health Organization, Off-label use of medicines for COVID-19

Pharmacologic treatments for coronavirus disease 2019 (covid-19): A review

Four ways researchers are responding to the covid-19 outbreak

Developing covid-19 vaccines at pandemic speed

Impact of non-pharmaceutical interventions (NPIs) to reduce covid-19 mortality and healthcare demand

The global impact of COVID-19 and strategies for mitigation and suppression

Comparing nonpharmaceutical interventions for containing emerging epidemics

When will the coronavirus outbreak peak?

Life has to go on': How sweden has faced the virus without a lockdown

The unusual ways countries are managing lockdowns

Hospitals in Brazil's So Paulo 'near collapse

Modeling infectious disease dynamics

Social Interaction Layers in Complex Networks for the Dynamical Epidemic Modeling of COVID-19 in Brazil

Modelling fatality curves of covid-19 and the effectiveness of intervention strategies

The global impact of COVID-19 and strategies for mitigation and suppression

Modeling the early evolution of the COVID-19 in Brazil: results from a Susceptible-Infectious-Quarantined-Recovered (SIQR) model (2020)

Modeling the epidemic dynamics and control of covid-19 outbreak in china

Feasibility of controlling COVID-19 outbreaks by isolation of cases and contacts, The Lancet Global Health

Emergence of scaling in random networks

The mathematics of infectious diseases

Statistical mechanics of complex networks

Modelling disease outbreaks in realistic urban social networks

Incubation period of 2019 novel coronavirus (2019-ncov) infections among travellers from wuhan, china

Estimating clinical severity of COVID-19 from the transmission dynamics in wuhan, china

novel coronavirus: Where we are and what we know

Optimal, near-optimal, and robust epidemic control

Incubation period and other epidemiological characteristics of 2019 novel coronavirus infections with right truncation: A statistical analysis of publicly available case data

Collective dynamics of 'small-world' networks