PLoS Comput Biol plos ploscomp PLOS Computational Biology 1553-734X 1553-7358 Public Library of Science San Francisco, CA USA PCOMPBIOL-D-23-01783 10.1371/journal.pcbi.1012182 Research Article Earth sciencesGeographyHuman geographyHuman mobility Social sciencesHuman geographyHuman mobility People and placesGeographical locationsEuropeEuropean UnionDenmark People and placesGeographical locationsEuropeEuropean UnionSweden People and placesGeographical locationsEuropeNorway Medicine and health sciencesEpidemiologyInfectious disease epidemiology Medicine and health sciencesMedical conditionsInfectious diseasesInfectious disease epidemiology People and placesGeographical locationsEuropeEuropean UnionFinland Medicine and health sciencesMedical conditionsInfectious diseasesViral diseasesCOVID 19 Medicine and health sciencesEpidemiologyPandemics The influence of cross-border mobility on the COVID-19 epidemic in Nordic countries The influence of cross-border mobility on the COVID-19 epidemic in Nordic countries https://orcid.org/0000-0002-0677-6761 Shubin Mikhail Conceptualization Formal analysis Investigation Methodology Software Validation Writing – original draft Writing – review & editing 1 * Brustad Hilde Kjelgaard Conceptualization Data curation Visualization Writing – original draft Writing – review & editing 2 https://orcid.org/0000-0002-8321-2724 Midtbø Jørgen Eriksson Data curation 3 Günther Felix Conceptualization Data curation Methodology Writing – original draft 4 Alessandretti Laura Data curation Software Writing – original draft Writing – review & editing 5 Ala-Nissila Tapio Conceptualization Funding acquisition Methodology Project administration Resources Writing – original draft Writing – review & editing 6 7 Scalia Tomba Gianpaolo Conceptualization Investigation Methodology Writing – original draft Writing – review & editing 4 8 Kivelä Mikko Conceptualization Funding acquisition Investigation Methodology Project administration Resources Supervision Writing – original draft Writing – review & editing 9 https://orcid.org/0000-0002-4689-6995 Chan Louis Yat Hin Writing – original draft Writing – review & editing 3 https://orcid.org/0000-0001-8411-8329 Leskelä Lasse Conceptualization Funding acquisition Investigation Methodology Project administration Resources Writing – original draft Writing – review & editing 1 Department of Mathematics and Systems Analysis, Aalto University, Espoo, Finland Oslo Center for Biostatistics and Epidemiology, Oslo University Hospital, Oslo, Norway Department of Method Development and Analytics, Norwegian Institute of Public Health, Oslo, Norway Department of Mathematics, Stockholm University, Stockholm, Sweden DTU Compute, Technical University of Denmark, Copenhagen, Denmark Quantum Technology Finland Center of Excellence, Department of Applied Physics, Aalto University, Espoo, Finland Interdisciplinary Centre for Mathematical Modelling and Department of Mathematical Sciences, Loughborough University, Loughborough, United Kingdom Department of Mathematics, University of Rome Tor Vergata, Rome, Italy Department of Computer Science, Aalto University, Espoo, Finland Struchiner Claudio José Editor Fundação Getúlio Vargas: Fundacao Getulio Vargas, BRAZIL

The authors have declared that no competing interests exist.

 * E-mail: mikhail.shubin@aalto.fi 6 2024 12 6 2024 20 6 e1012182 2 11 2023 20 5 2024 2024 Shubin et al This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Restrictions of cross-border mobility are typically used to prevent an emerging disease from entering a country in order to slow down its spread. However, such interventions can come with a significant societal cost and should thus be based on careful analysis and quantitative understanding on their effects. To this end, we model the influence of cross-border mobility on the spread of COVID-19 during 2020 in the neighbouring Nordic countries of Denmark, Finland, Norway and Sweden. We investigate the immediate impact of cross-border travel on disease spread and employ counterfactual scenarios to explore the cumulative effects of introducing additional infected individuals into a population during the ongoing epidemic. Our results indicate that the effect of inter-country mobility on epidemic growth is non-negligible essentially when there is sizeable mobility from a high prevalence country or countries to a low prevalence one. Our findings underscore the critical importance of accurate data and models on both epidemic progression and travel patterns in informing decisions related to inter-country mobility restrictions.

 Author summary

A typical intervention during pandemics such as COVID-19 is to restrict the mobility of individuals and thus prevent or slow down the spreading process. The role of within-country mobility has been studied in several countries, but the role of inter-country mobility is less well understood. To assess the effects of border closures that may cause significant economic and societal harm, it is necessary to understand in detail their efficacy from an epidemiological point of view. In the present work we model the effect of inter-country mobility on the spread of COVID-19 during 2020 in the neighbouring Nordic countries of Denmark, Finland, Norway and Sweden. We investigate the immediate impact of cross-border travel and employ counterfactual scenarios to explore the cumulative effects of introducing additional infected individuals into a population during the ongoing epidemic. Our results suggest that the effect of inter-country mobility on epidemic growth is non-negligible essentially when there is sizeable mobility from a high prevalence country or countries to a low prevalence one.

 http://dx.doi.org/10.13039/501100004785 Nordforsk 105572 NordicMathCovid This work has been funded in part by the project 105572 NordicMathCovid as part of the Nordic Programme on Health and Welfare funded by NordForsk. The funder had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. PLOS Publication Stage vor-update-to-uncorrected-proof Publication Update 2024-06-25 Data Availability All relevant data and code are published separately at https://gitlab.com/2pi360/covid_model_mobility_public. Outbreaks COVID-19 1 Introduction

Mobility and the consequent human contacts are the key factors for the spread of infectious diseases. Governments frequently impose and enforce border restrictions on individuals to prevent a disease from entering a country or to slow its spreading. However, such interventions can be problematic for economic, legal and social reasons. To assess the effects of border restrictions, quantitative modeling and prediction of their efficacy should be provided. This topic has experienced renewed interest each time a pandemic threat has emerged e.g., avian flu, SARS-CoV-1, Ebola and, most recently, COVID-19. Various approaches are possible, including descriptive studies [1], studies based on genomic analyses [2–4], and studies investigating social and health aspects [5]. However, the variety of mobility restriction schemes enacted by different countries during the COVID-19 pandemic and the general lack of quantitative evaluation of their differences and effects indicate a need for more analytical studies (e.g., [6, 7]). This is particularly important as restrictions on cross-border mobility may have serious detrimental effects on the economy and thus need to be proven to be effective on short and long time scales [8].

Most of the studies to date focus on the use of border restrictions in avoiding or delaying the start of a local epidemic. It has been shown (see [9] and references therein) that, if the restrictions are not 100% effective, they only postpone the start of the epidemic, and the delay time gained is usually short. Furthermore, restrictions must be enacted before the infection has been introduced in the country, since local spread dynamics will quickly dominate over importation. If infections are added in a later phase of the epidemic, the effect may be small because individuals that would be infected by the imported infection could be infected in any case due to the local epidemic [10].

Our interest in the present work is in the quantitative estimation of the effects of cross-border traffic on the number of infected in the Nordic countries sharing a common border during an ongoing epidemic. The COVID-19 pandemic in the Nordic countries during 2020 provides an interesting case study for the effect of border crossing traffic. First, the Nordic countries have a had a free movement agreement since 1954, but various exceptions occurred during the COVID-19 period. Second, there were major differences between the internal restrictions and the overall strategies carried out in each country, which led to large differences in the scale and timeline of the pandemic. While movement restrictions within these countries during the epidemic have been analysed, the analyses seem mostly to focus on legal or social aspects (see, e.g., [11, 12]). There are also analyses of the different intervention strategies and their effects in these countries, but without explicit analysis of the interactions between the countries [13].

The interaction between cross-border traffic and epidemic spread can be analysed in several ways, e.g., how epidemic outbreaks affect mobility (see, e.g., [6, 14, 15]) or vice versa, how mobility affects epidemic spread. For designing and enforcing effective but minimally disruptive restrictions on mobility and travel, the latter aspect is more important. Such studies can be both retrospective [10, 16], i.e., based on available data on traffic and disease, or prospective [17], i.e., entirely model-based. In both cases, however, the effects will usually be evaluated by comparing scenarios with different amounts of traffic. In the retrospective case, this will involve counterfactual scenarios [13]. There are two main problems with such an approach. One is the need to specify how societies respond to new disease scenarios and the other is the choice of evaluation criteria of the differences between scenarios. Consider, for example, the term “first infective” (also sometimes referred as index case) from whom a local epidemic originates; this case is usually considered to be imported. Should we define the effect of importing this infection as the final size of the whole epidemic or just the individuals directly infected by that case? What if it were assumed that the infection would have been introduced anyway, even if the first infective had failed to infect anyone? It is thus important to clearly define measures of difference between scenarios and to distinguish the effects of internal and external forces of infection and of primary (direct) and secondary (consequential) effects of new infections into the population.

Our approach to evaluating the effects of inter-country mobility is to focus on two extreme cases: 1) Primary effects of mobility measured by descriptive statistics, i.e., the number of infected individuals that travel to and from each country. 2) Secondary effects, which we compute by running various counterfactual scenarios where we eliminate or restore cross-border traffic to pre-epidemic levels, but keep everything else unchanged, and simulate epidemic spreading under these new conditions. This is likely going to overestimate the number of infected over longer time intervals as large increases in the number of infections would probably have been met with stricter restrictions and changes in population behaviour. We thus present the effects over reasonable short forward time intervals. While neither of these evaluation methods give a realistic picture of the effects of mobility restrictions, they may serve as upper and lower bounds for the effects, and taken together they serve as a useful tool for assessing the range of possible outcomes. Our results indicate that the effect of inter-country mobility on epidemic growth is non-negligible essentially when there is sizeable mobility from a high prevalence country or countries to a low prevalence one.

To quantitatively assess the problem at hand we formulate an SIR mathematical model (representing the numbers of Susceptible, Infectious and Recovered individuals and their evolution over time in the Nordic countries; also known as an Eulerian approach) that explicitly uses estimates of cross-border movements of individuals. These movements are subdivided into short-term (commuter) and long-term visits in the receiving country. Combined with hospitalisation data from the modelled countries (and estimates of numbers of imported infections from the rest of the world), we infer the numbers of imported cases and their effects on within-country reproductive numbers during different phases of the disease spread. The model is then applied to the first year of the COVID-19 pandemic in the four Nordic countries sharing common borders: Denmark, Finland, Norway and Sweden. Traffic between the Nordic countries and the rest of the world is thus considered in the model.

The structure of this paper is as follows: In Section 2 we discuss the data and models used in the present study. Section 3 presents the measures and results for the primary effects, i.e., the direct impact of mobility. Section 4 presents the secondary effects of mobility, i.e., the counterfactual scenarios and their results. Finally, Section 5 summarises and discusses the conclusions of our study.

 2 Data and models

Our study requires accurate and detailed data on the course of the epidemic within the countries, mobility between them, and a model that can be used to represent these data. Section 2.1 introduces the health data and Section 2.2 and S1 Appendix cover the mobility data. Section 2.3 describes the model (See also S2 Appendix for the derivation of the model).

2.1 Health data

To calibrate our model, we used the weekly numbers of new hospitalisations due to COVID-19. Data were obtained from the Finnish Institute for Health and Welfare (Finland), Socialstyrelsen [18] (Sweden), the Norwegian Institute of Public Health [19] (Denmark and Norway) and Our World in Data website [20] (rest of the world). We used aggregated data without age-based or regional stratification.

 2.2 Mobility data

A central part of the research project has been the quantification of the number of individuals moving between the four bordering Nordic countries. To this end, we collected a data set consisting of 12 directed passenger flows between the Nordic countries Denmark (population 5.9 million people), Finland (5.5 M), Norway (5.4 M) and Sweden (10.5 M) and 4 directed passenger flows from the rest of the world into the 4 countries, and 4 directed passenger flows from the 4 countries to the rest of the world. The transmission model used to simulate the epidemic for each country has a daily temporal resolution. Hence, a daily temporal resolution is needed for the mobility data as well.

The directed flows represent the daily number of individuals travelling by air, road, railway and ferry. Fig 1A visualizes the temporally aggregated mobility between the countries in our data. We note that not all modes of transportation are relevant for all combinations of countries, e.g., no direct railway connection exists between Denmark and Norway. S1 Appendix presents more details on the collection of data for the different sources of transportation for the different countries.

10.1371/journal.pcbi.1012182.g001 Mobility between the four Nordic and the other countries, represented by lines connecting the origin and destination.

The widths of the lines are proportional to the numbers of border crossings during the modeling period (10 February—31 December 2020). (A): Line color represents the transportation type as indicated in the legend. Numbers near the lines show the total number of border crossings in millions (M). For comparison, populations of Denmark, Finland and Norway are about 5.5 M each, while that of Sweden is 10.5 M. (B) The number of commuters and long-term travellers (C) Schematic of commuters, long-term travellers and their representation in the model. Left sub-panel shows the base case scenario without any travel. Middle sub-panel shows an example of long-duration travel—an infected individual moves from country A to country B and stays there; this move creates extra infection pressure in B and reduces infection rate in country A. Right sub-panel shows an example of a commuter from country A getting infected in country B and continuing to move between countries.

We split the passenger flows into two categories: commuters and long-term travellers. We define commuters as one-day travellers, meaning individuals that exit a country and return to the same country the same day. The most probable transportation methods for commuters are by road. The road traffic data we have collected have an hourly temporal resolution. By modelling the road traffic, i.e., the time point of each border crossing, by a Gaussian mixture model (GMM), we identify components (Gaussian distributions) of the fitted model that correspond to commuting based on the estimated mean value (corresponding to a time point during the day) of the components. The commuting components then give the probability that a travel occurring at a specific time of the day is of commuter type. For more details, see S1 Appendix. Commuting also occurs by railway between Denmark and Sweden across the Öresund bridge. However, the temporal resolution of the railway data is not accurate enough to apply a GMM. We therefore assume the same fraction of commuters for railway traffic across the Öresund Bridge as for road traffic across it. Fig 1B visualizes how the mobility data are split into commuters and long-term travelers. The widths of the lines are proportional to the number of travels during the modelling period.

We collected data for the entire years of 2019 and 2020. The modelling period is the time interval 10 February 2020—31 December 2020. We used data for pre-pandemic year 2019 as a reference for the counterfactual scenarios to be discussed below.

 2.3 Model

We use an extended Susceptible-Infected-Removed (SIR) model that tracks the epidemic status in each country. In our model time is discretised with a time step equal to one day. The model has three components: a deterministic transmission model simulates the epidemic, an observation model links the simulated number of infections to the observed data, and a parameter model defines the parameters for transmission and observation models. Fig 2 presents a graphical overview of the model while Table 1 summarizes the notation.

10.1371/journal.pcbi.1012182.g002 Graphical representation of the SIR model adopted for the present study.

The model is spatially resolved at the level of each Nordic country and incorporates flows of travelers between them. The model part (transmission, observation, parameter) is indicated by the color. The schematic indicates both the model variables and data for hospitalisation and mobility. For more details, see Table 1 for notation and Section 2 on various components of the model.

 10.1371/journal.pcbi.1012182.t001 Notation glossary.
Parameter Value Definition
Transmission model
R t , x estimated Time- and country-dependent reproduction number. This number aggregates the factors influencing the spread of infection: interventions, changes in population behaviour, seasonal effects and super-spreading outbreaks.
St,x, It,x, Rt,x estimated Numbers of Susceptible, Infectious and Removed individuals per day per country
Nt,x St,x + It,x + Rt,x Population size
γ 1/8 (per day) Recovery rate
i t , x inflow given External infectious flow, i.e., number of infectious individuals arriving to country x on day t from outside of the Nordic countries
i t , x new Eq 4 The number of residents of country x infected on day t
rt,x Eq 5 Risk of infection in country x on day t, rt,x ∈ [0, 1]
D t , d , s com , D t , d , s long given Counts of commuter and long-term travellers flow, i.e., the number of individuals going from country s to d on day t
M t , d , s com , M t , d , s long Eqs 6 and 7 Population-relative mobility matrices
M t , o , s long given Population-relative outflow, i.e., the portion of population leaving country s on the day t to non-Nordic countries
αd,s 0.5 Average fraction of time that a commuter from country s spends in country d during a daily trip
Observation model
Et,x Eq 8 Expected numbers of hospitalisations
Ht,x given Observed number of hospitalisations
hx 2% The hospitalisation probability
pt given The probability that hospitalisation happens t days after infection
2.3.1 Transmission model

The transmission model is spatially resolved at the level of each country and incorporates two types of mobilities: daily commuters and long-term travellers. We use an Eulerian approach where we keep track of individuals who are currently present in region x, and we follow the long-term flows of individuals across the regions [21]. In addition, we incorporate a Lagrangian model component to capture the effect of daily commuters as in [22]. We denote by St,x, It,x, and Rt,x the expected numbers of susceptible, infectious, and removed individuals who are in the beginning of day t ∈ {0, 1, 2, …} located in country x ∈ {1, 2, 3, 4}. We allow population sizes of countries changes with time: Nt,x = St,x + It,x + Rt,x ≠ const. Transmission characteristics are parameterised by the recovery rate γ = 1/8 (per day) and time-dependent country-specific reproduction rate R t , x > 0. Mobility is parameterised using external infectious flow counts i t , x inflow, commuter flow counts D t , x , y com, and long-term flow counts D t , x , y long indexed by t ≥ 1, and x ≠ y.

The model evolves according to S t + 1 , x= ∑ y M t , x , y long S t , y - i t , x new ; I t + 1 , x= ∑ y M t , x , y long I t , y + i t , x new + i t , x inflow - I t , x M t , o , x long - γ I t , x ; R t + 1 , x= ∑ y M t , x , y long R t , y + γ I t , x , where i t , x new = S t , x ∑ y M t , y , x com r t , y is the number of residents of country x infected on day t, and where r t , y = R t , y γ ∑ z M t , y , z com I t , z / N t , y represents the risk of infection in country y on day t. Here M t , d , s com and M t , d , s long are population-relative mobility matrices, defined as follows. Let D t , d , s com and D t , d , s long be the number of people from the source country s travelling to the destination country d for a short (i.e., less than one day) or a long (i.e., one day or longer) visit, respectively. We define the relative matrices as: M t , d , s long= { D t , d , s long N t , s , for s ≠ d ; 1 - ∑ x ≠ s D t , x , s long N t , s , for s = d ; M t , d , s com= { α d , s D t , d , s com N t , s , for s ≠ d ; 1 - α d , s ∑ x ≠ s D t , x , s com N t , s , for s = d , where αd,s = 0.5 is a factor representing the average fraction of time that a commuter from country s spends in country d during a daily trip.

The model is initialized on day t = 0 corresponding to 10 February 2020, with 0.01% of the population being infected, i.e., such that S 0 , x = 0 . 9999 N x ; I 0 , x = 0 . 0001 N x ; R 0 , x = 0 , where Nx is the population size of country x. Details of the derivation of the transmission model and its parameters are presented in S2 Appendix.

 2.3.2 Observation model

We define the expected number of hospitalisations on day t in country x as a convolution of the number of new infections: E t , x = ∑ τ = 0 t - 1 h x p t - τ i t , x new , where hx = 0.02 is the hospitalisation probability in country x and pd is the probability that hospitalisation happens d days after infection. We set pd to be a negative binomial distribution with mean of 11 days and standard deviation of 5.

Let Hw,x be the number of observed hospitalisations in country x on week w. We link it to the expected numbers with the Negative Binomial (NB) distribution. H w , x ∼ NB ( mean = ∑ t = 7 w 7 w + 6 E t , x , overdispersion = 0 . 1 ) .

2.3.3 Parameter model and Bayesian parameter inference

The time-dependent reproduction numbers R t , x are the only free parameters in our model. This parameter is supposed to aggregate the factors influencing the spread of infection: interventions, changes in population behaviour, seasonal effects and super-spreading outbreaks. Our goal is to fit the reproduction numbers R given the available data: number of hospitalisations H, inflow iinflow and mobility matrices Dcom, Dlong and M • , 0 , • long. For each country, we estimate the posterior distribution P ( R | H , D com , D long , i inflow , M • , 0 , • long ) (where R represents a vector { R t , x , t ≥ 0 } ) using an adaptive Markov chain Monte Carlo (MCMC) algorithm. We allow the chain to run for 15,000 warm-up iterations, and then we run it for 300,000 iterations recording every 10th. On each iteration R x corresponding to each country is updated separately.

We construct the prior for R x as a Gaussian random walk. The initial reproduction number is a priori sampled from a truncated normal distribution, limited to positive values only such that R t = 0 , x ∼ Normal + ( 2 , 0 . 5 ). The reproduction number on each subsequent Monday (t > 0, t mod 7 = 0) is sampled as R t , x ∼ Normal + ( R t - 7 , x , 0 . 25 ). The rest of the values are linearly interpolated between the closest Mondays. Proposal distributions for all batches of parameters are Multivariate Normal, with covariance adapted after each iteration. The details of the adaptation code are presented in [23] and the code implementing our model and fitting is available at gitlab.com/2pi360/covid_model_mobility_public.

 3 Primary effects: Direct effects of imported and exported cases 3.1 Estimates of the effect of the mobility

This section defines the metrics we use to evaluate the effects of mobility. As the early data on the inflow of infections i t , x inflow are unreliable, we only show these estimates for dates starting at April (t = 51).

3.1.1 Net effect of mobility

The right side of Eq 2 can be written as I t , x ( 1 - γ ) ︸ unrecovered existing infections + i t , x new ︸ ( A )new infections among residents of x + ∑ y , y ≠ x M t , x , y long I t , y ︸ ( B )infected from other Nordics arriving to x- ∑ y , y ≠ x M t , y , x long I t , x ︸ ( C )departing to other Nordic countries from x + i t , x inflow ︸ ( D )inflow from outside of Nordics to x - M t , o , x long I t , x ︸ ( E )departing from x to outside of Nordics . Here the term A refers to the new infections which occurred among the residents of country x, B to infections coming to x from other Nordic countries, C to infections leaving from x to other Nordic countries, D to new infections coming to x from non-Nordic countries, and E to infections leaving from x to non-Nordic countries. All the quantities A, B, C, D and E are positive. We refer to Q t , x long term = B - C as the net flow due to long-term travel into x and Q t , x non-Nordic = D - E as the net flow from non-Nordic countries into x. The quantity i t , xtotalnew = i t , x new + Q t , xlongterm + Q t , xnon-Nordic= A + B - C + D - E is referred to as the total number of new infections. Terms B and C contain contributions from the individual Nordic countries. Term A defined in Eq 11 can be further written as: i t , x new = R t , x γ N t , x S t , x I t , x ︸ (F )counterfactual local infections - R t , x γ N t , x S t , x I t , x ( 1 - ( M t , x , x com ) 2 ) ︸ ( G )reduction in local infections due to commuters leaving + R t , x γ N t , x S t , x M t , x , x com ∑ y , y ≠ xM t , x , y com I t , x ︸ ( H)infections caused by commuters arriving from other Nordics to country x + ∑ y , y ≠ xR t , y γ N t , y S t , x M t , y , x com ∑ zM t , y , z com I t , z ︸ (J)commuters from country x , infected in other countries and returning back .

The quantities F, G, H and J are positive. Term F refers to the counterfactual number of the local infections in country x which would have occurred if no commuters had either arrived or left the country. Term G refers to the number of infections inside country x prevented by commuters leaving the country (this decreases both infection pressure and the pool of susceptibles). Term H refers to the infections caused by arriving commuters in the local population. Finally, term J represents the commuters from country x who got infected while abroad and returned back to x. We refer to the term i t , x local = F as local infections and to Q t , x commuting = H + J - G as the net commuting effect. We further refer to the sum Q t , x = Q t , xlongterm+ Q t , xnon-Nordic+ Q t , x commuters = B - C + D - E - G + H + J as the net mobility effect. In the following, we also define the relative net mobility effect as a net mobility effect divided by the total number of new infections, i.e., i t , xnewrel= Q t , x i t , xtotalnew = B - C + D - E - G + H + J A + B - C + D - E .

3.1.2 Effective reproduction and multiplication numbers

One of the main metrics of epidemic growth, the effective reproduction number R t , x eff, is often defined as the average number of secondary infections caused by a single infected individual who got infected at day t in country x. This definition describes local transmission and is not suitable for our purposes. We thus suggest an effective multiplication number M t , x eff as the number of new infections that emerged in country x for any reason during the average infectious period, divided by the number of new infections at day t.

Our definition of the parameter R already implicitly includes the effects from intervention, behavioral changes, weather etc., so the difference between R t and R t eff can only be caused by reduction in the susceptible population. R t , x eff can be approximated as R t , x S t , x / N t , x, where St,x/Nt,x is a remaining fraction of susceptibles. Alternatively, it can be approximated as a number of new infections per each infectious person on day t, multiplied by the average number of days one stays infectious 1/γ. The number of new local infections in the absence of mobility is denoted as i τ , xtotalnew (see Eq 13), and thus R t , x eff ≈ R t , x S t , x N t , x = i t , x local I t , x γ . In analogy to the equation above, we approximate M t , x by using the total number of new infections instead of only local ones: M t , x eff = i τ , xtotalnew I t , x γ . Given that R is defined as a metric of epidemic growth due to the local transmission and M is a metric of growth due to all factors, M - R represents the growth contributed by mobility. This kind of distinction between sources of epidemic growth has recently been made between the effects of local versus imported cases (see, e.g. [10, 24, 25]), although the exact definition of importation varies.

 3.1.3 Prevalence of infection

The prevalence of local infections in country x is defined as πt,x = It,x/Nx; the prevalences of infection among commuters and among long-term travellers from Nordic countries, arriving to country x, are defined as π t , x com = ∑ y , y ≠ x I t , x D t , x , y com / N t , y ∑ y , y ≠ x D t , x , y com , and π t , x long = ∑ y , y ≠ x I t , x D t , x , y long / N t , y ∑ y , y ≠ x D t , x , y long , respectively. Prevalence of infection among long-term travellers from outside Nordic countries π t , x inflow computed directly from the data, see S2 Appendix for details.

 3.2 Results

We will next describe the results of applying our approach to the four neighboring Nordic countries during April–December 2020. The progress of the epidemic differs between the countries (see Fig 3, also S3 Appendix Fig F). Denmark, Finland and Norway initially followed the same epidemic trajectory: in early April the incidence rate, which we always give as new infections per day per 10,000 inhabitants, was about 5. This incidence rate quickly decreased and stayed below 1 − 2 until October. On the other hand, Sweden had an incidence rate of 20 during the whole of April and this rate dropped to 1 only by July. In October–December Finland, Norway and Sweden experienced a second wave which reached incidence rates of 3 for Finland and Norway and 10 for Sweden. In Denmark, the second wave started earlier, in September and reached an incidence rate of 40 by December.

10.1371/journal.pcbi.1012182.g003 <p>Black color: estimates of the total number of new infections <inline-formula id="pcbi.1012182.e062"><alternatives><graphic id="pcbi.1012182.e062g" mimetype="image" position="anchor" xlink:href="info:doi/10.1371/journal.pcbi.1012182.e062" xlink:type="simple"/><mml:math display="inline" id="M62"><mml:msubsup><mml:mi>i</mml:mi> <mml:mrow><mml:mi>t</mml:mi> <mml:mo>,</mml:mo> <mml:mi>x</mml:mi></mml:mrow><mml:mrow><mml:mtext>total</mml:mtext><mml:mspace width="2pt"/><mml:mtext>new</mml:mtext></mml:mrow></mml:msubsup></mml:math></alternatives></inline-formula> of <xref ref-type="disp-formula" rid="pcbi.1012182.e040">Eq 13</xref> per day per 10,000 individuals (lines show posterior mean and colored areas show 90% posterior intervals). Orange: estimates of the relative net mobility effect of <xref ref-type="disp-formula" rid="pcbi.1012182.e044">Eq 15</xref> <inline-formula id="pcbi.1012182.e063"><alternatives><graphic id="pcbi.1012182.e063g" mimetype="image" position="anchor" xlink:href="info:doi/10.1371/journal.pcbi.1012182.e063" xlink:type="simple"/><mml:math display="inline" id="M63"><mml:msubsup><mml:mi>i</mml:mi> <mml:mrow><mml:mi>t</mml:mi> <mml:mo>,</mml:mo> <mml:mi>x</mml:mi></mml:mrow><mml:mrow><mml:mtext>new</mml:mtext><mml:mspace width="2pt"/><mml:mtext>rel</mml:mtext></mml:mrow></mml:msubsup></mml:math></alternatives></inline-formula>, defined by dividing the net mobility effect (cf. <xref ref-type="disp-formula" rid="pcbi.1012182.e043">Eq 14</xref>) by the total number of new infections (lines show posterior mean and colored areas show 90% posterior intervals). Note that the net mobility effect can be negative, meaning that the country is a net exporter of infections.</p> </caption> <graphic mimetype="image" position="float" xlink:href="info:doi/10.1371/journal.pcbi.1012182.g003" xlink:type="simple"/> </fig> <p>The first interesting result from our modeling study is that during the time period considered here, the net effect of inter-country mobility is very small as compared to the number of local infections. To quantify its effect, we computed the <italic>relative net mobility effect</italic> <inline-formula id="pcbi.1012182.e064"><alternatives><graphic id="pcbi.1012182.e064g" mimetype="image" position="anchor" xlink:href="info:doi/10.1371/journal.pcbi.1012182.e064" xlink:type="simple"/><mml:math display="inline" id="M64"><mml:mrow><mml:msubsup><mml:mi>i</mml:mi> <mml:mrow><mml:mi>t</mml:mi> <mml:mo>,</mml:mo> <mml:mi>x</mml:mi></mml:mrow> <mml:mrow><mml:mtext>new</mml:mtext><mml:mspace width="2pt"/><mml:mtext>rel</mml:mtext></mml:mrow></mml:msubsup> <mml:mo>=</mml:mo> <mml:msub><mml:mi>Q</mml:mi> <mml:mrow><mml:mi>t</mml:mi> <mml:mo>,</mml:mo> <mml:mi>x</mml:mi></mml:mrow></mml:msub> <mml:mo>/</mml:mo> <mml:msubsup><mml:mi>i</mml:mi> <mml:mrow><mml:mi>t</mml:mi> <mml:mo>,</mml:mo> <mml:mi>x</mml:mi></mml:mrow> <mml:mrow><mml:mtext>total</mml:mtext><mml:mspace width="2pt"/><mml:mtext>new</mml:mtext></mml:mrow></mml:msubsup></mml:mrow></mml:math></alternatives></inline-formula> (see <xref ref-type="disp-formula" rid="pcbi.1012182.e044">Eq 15</xref>). As seen in <xref ref-type="fig" rid="pcbi.1012182.g003">Fig 3</xref>, this fraction is close to zero in Finland almost until mid-May. The relative net mobility effect then peaks around June–July, reaching 40% (0.04 out of the total 0.1 incidence rate is explained by mobility). Norway exhibits a similar pattern to Finland, with the peak time starting earlier already in April, stretching slightly longer in time and reaching 30% (0.1 out of the total 0.3 incidence rate is explained by mobility). Denmark displays a similar, but even less pronounced, peak during spring and summer, reaching 4% (0.002 out of the total 0.05 incidence rate is explained by mobility). However, Denmark becomes a net exporter of infections in August, which means that there were more infected individuals travelling out of the country than coming in during this time period. Sweden on the other hand is a net exporter of infections in our model until late August, however <inline-formula id="pcbi.1012182.e065"><alternatives><graphic id="pcbi.1012182.e065g" mimetype="image" position="anchor" xlink:href="info:doi/10.1371/journal.pcbi.1012182.e065" xlink:type="simple"/><mml:math display="inline" id="M65"><mml:msubsup><mml:mi>i</mml:mi> <mml:mrow><mml:mi>t</mml:mi> <mml:mo>,</mml:mo> <mml:mi>x</mml:mi></mml:mrow> <mml:mrow><mml:mtext>new</mml:mtext><mml:mspace width="2pt"/><mml:mtext>rel</mml:mtext></mml:mrow></mml:msubsup></mml:math></alternatives></inline-formula> never exceeds 10%.</p> <p>Our model shows that for the inter-country mobility to have a non-negligible effect, three conditions have to be met: (i) the number of new infections in the target country must be very low, (ii) the number of infections in the source country must be large compared to the target country, and (iii) the mobility between the countries must be sufficiently large to transfer sizable amounts of infection. There are two time periods when the first two conditions were met, namely April–August 2020 when Sweden had significantly more new infections than the other countries, and December 2020 when Denmark had a high incidence (see <xref ref-type="fig" rid="pcbi.1012182.g003">Fig 3</xref>). While difference in incidence between two countries is necessary for border crossings to make a difference in our model, it is not a sufficient condition, because the rate of border crossings might not be large enough to enhance the rate of infections. This is the case in the two aforementioned instances, as will be discussed below.</p> <p>We next turn to the question of what type of travellers brought in the infections, and use the division of border crossings into three groups described in Section 2.2: commuters, long-term travellers from the four Nordic countries, and travellers from the rest of the world. <xref ref-type="fig" rid="pcbi.1012182.g004">Fig 4</xref> shows the effect of the estimated total net flow of infected travellers on the number of infections per day per 10,000 using this decomposition. The contribution of individuals identified as commuters is consistently very low in our model. This is due to a combination of two effects: the relatively low number of commuters and the fact that the model yields a smaller effect per border crossing for commuters because they spend only part of the day in the country whereas long-term travellers are assumed to reside in the country for their remaining infectious period. See <xref ref-type="supplementary-material" rid="pcbi.1012182.s003">S3 Appendix</xref> Fig G for the relative numbers and Fig H for the further split of flows.</p> <fig id="pcbi.1012182.g004" position="float"> <object-id pub-id-type="doi">10.1371/journal.pcbi.1012182.g004</object-id> <label>Fig 4</label> <caption> <title>Results of the net mobility effect on the number of infections per day per 10,000 caused by different modes of travel: <inline-formula id="pcbi.1012182.e066"><alternatives><graphic id="pcbi.1012182.e066g" mimetype="image" position="anchor" xlink:href="info:doi/10.1371/journal.pcbi.1012182.e066" xlink:type="simple"/><mml:math display="inline" id="M66"><mml:msubsup><mml:mi>Q</mml:mi> <mml:mrow><mml:mi>t</mml:mi> <mml:mo>,</mml:mo> <mml:mi>x</mml:mi></mml:mrow> <mml:mrow><mml:mtext>long</mml:mtext><mml:mspace width="2pt"/><mml:mtext>term</mml:mtext></mml:mrow></mml:msubsup></mml:math></alternatives></inline-formula>, <inline-formula id="pcbi.1012182.e067"><alternatives><graphic id="pcbi.1012182.e067g" mimetype="image" position="anchor" xlink:href="info:doi/10.1371/journal.pcbi.1012182.e067" xlink:type="simple"/><mml:math display="inline" id="M67"><mml:mrow><mml:msubsup><mml:mi>Q</mml:mi> <mml:mrow><mml:mi>t</mml:mi> <mml:mo>,</mml:mo> <mml:mi>x</mml:mi></mml:mrow><mml:mrow><mml:mtext>non-Nordic</mml:mtext></mml:mrow></mml:msubsup></mml:mrow></mml:math></alternatives></inline-formula> and <inline-formula id="pcbi.1012182.e068"><alternatives><graphic id="pcbi.1012182.e068g" mimetype="image" position="anchor" xlink:href="info:doi/10.1371/journal.pcbi.1012182.e068" xlink:type="simple"/><mml:math display="inline" id="M68"><mml:msubsup><mml:mi>Q</mml:mi> <mml:mrow><mml:mi>t</mml:mi> <mml:mo>,</mml:mo> <mml:mi>x</mml:mi></mml:mrow> <mml:mtext>commuters</mml:mtext></mml:msubsup></mml:math></alternatives></inline-formula> (see text for details).

Lines show posterior mean and colored areas show 90% posterior intervals.

The data on incidences in Fig 3 and on the traveller types of Fig 4 suggest that the significant increase in the relative number of imported infections during May—August 2020 in Finland, Norway, and Denmark was due to infected travelers from Sweden. This is confirmed by Fig 5, which exhibits the effect of the countrywise net flows on the number of infections per day per 10,000 (refer to S3 Appendix Fig I for relative numbers). Note, however, that in December 2020, and to some extent already starting from September, Denmark became a net contributor of infections to the other Nordic countries.

10.1371/journal.pcbi.1012182.g005 Daily net mobility effect on the number of infections per day per 10,000 individuals from each source country.

The color legends are in the upper right panel. The lines show posterior mean and colored areas show 90% posterior intervals.

The results shown here carry somewhat contradictory messages. One one hand, they show that the net effect of the inter-country mobility is very low. On the other hand, when local transmission rate is low, incoming infections may play a substantial part in the epidemic. To clarify the message, we can examine the epidemic trajectories from another angle, namely by comparing the effective reproduction and multiplication numbers R eff and M eff, respectively (Fig 6). When M eff > R eff mobility introduces extra infections into the population, but when M eff < R eff mobility removes infections from the population. In particular, when M eff > 1 and R eff < 1 the epidemic can only grow because of the inter-country mobility. Such a situation occurred for Finland and Norway during June—August 2020.

10.1371/journal.pcbi.1012182.g006 Estimates of daily effective reproduction number <inline-formula id="pcbi.1012182.e075"><alternatives><graphic id="pcbi.1012182.e075g" mimetype="image" position="anchor" xlink:href="info:doi/10.1371/journal.pcbi.1012182.e075" xlink:type="simple"/><mml:math display="inline" id="M75"><mml:msubsup><mml:mi mathvariant="script">R</mml:mi> <mml:mrow><mml:mi>t</mml:mi> <mml:mo>,</mml:mo> <mml:mi>x</mml:mi></mml:mrow> <mml:mtext>eff</mml:mtext></mml:msubsup></mml:math></alternatives></inline-formula> and multiplication number <inline-formula id="pcbi.1012182.e076"><alternatives><graphic id="pcbi.1012182.e076g" mimetype="image" position="anchor" xlink:href="info:doi/10.1371/journal.pcbi.1012182.e076" xlink:type="simple"/><mml:math display="inline" id="M76"><mml:msubsup><mml:mi mathvariant="script">M</mml:mi> <mml:mrow><mml:mi>t</mml:mi> <mml:mo>,</mml:mo> <mml:mi>x</mml:mi></mml:mrow> <mml:mtext>eff</mml:mtext></mml:msubsup></mml:math></alternatives></inline-formula>, as approximated in Eqs <xref ref-type="disp-formula" rid="pcbi.1012182.e053">16</xref> and <xref ref-type="disp-formula" rid="pcbi.1012182.e055">17</xref>.

The lines show posterior mean and colored areas show 90% posterior intervals.

While the analysis above tells us about the importance of the border crossings for the epidemic pressure on the Nordic countries, it does not directly tell us about the effectiveness of interventions targeting traffic across borders. Here, instead of asking the question whether of not interventions should be implemented, we want to answer the question of whether it would be more effective to enforce interventions on the local population or on the population crossing the borders. To this end, we compute the prevalence, i.e., the proportion of infectious individuals, among the local population (Ia/Na) and the long-term travellers arriving to the country from Nordics and non-Nordic countries (see Fig 7). There are very large differences in the prevalences of local and travelling populations. Most notably, in Norway and Finland the Nordic travellers are more than one hundred times more likely to be infectious than the local population during the May—August of 2020. This indicates that every test done at the border during that period (for individuals without symptoms and no knowledge of exposure) was potentially more than a hundred times more efficient than tests for the local populations. Similarly, offsetting the effect of a single border crossing could potentially require much larger local movement restrictions. The opposite is true for Sweden, where until mid August, the prevalence of the local population is larger than the prevalence of Nordic travellers.

10.1371/journal.pcbi.1012182.g007 Prevalence among the local population for different travel types, and for long-term travellers visiting a country.

Lines show posterior mean and colored areas show 90% posterior intervals.

 4 Secondary effects: Counterfactual scenarios

The previous section focused on the number of infections coming into a given country. However, such numbers might underestimate the impact of border crossings: imported infections can lead to further infections, which are then classified as local infections as they are taking place within the country. These further infections could have been (partly) prevented if the original case in the infection chain were prevented. We use counterfactual scenarios to investigate the effect of full infection chains caused by border traffic. The border effect is isolated by changing the rate of mobility while keeping everything else in the model (including the disease spreading parameters) constant.

The counterfactual scenarios are not necessarily accurate predictions of what would have happened if travel restrictions or testing policies at the border were changed. There are two main reasons: firstly, changing policies at borders can alter the behaviour within the country in multiple ways. Some people may take border closure as a signal from authorities of the seriousness of the epidemic and abstain from all types of contacts. Others might instead choose to travel more within the country. Secondly, the within-country policies and behaviour are affected by the current epidemic situation, but in our counterfactual scenarios we assume that the country-specific reproduction rates R t , x remain unchanged. For example, if the overall disease burden went down due to restricted cross-border traffic, this could imply that people mix more inside a country and thereby increase the reproduction rate. These responses are difficult to model due to the general nonlinearities and lack of monotonicity inherent in the transmission dynamics [26].

Since we do not model the societal responses to the current epidemic situation, we expect the reliability of our estimates to go down for longer time horizons. Nevertheless, our estimates should serve as an upper bound for the change in number of infected individuals due to border traffic when this traffic increases the number of infections, because more infections would likely result in a reaction decreasing the local reproduction rates rather than increasing them. Similarly, if the changes in border traffic caused a lower number of infections within the country, the counterfactual would serve as a lower bound for the effect as lower number of infected people would likely lead to unchanged or higher reproduction numbers. Also, we do not investigate realistic scenarios where one would slightly adjust the numbers of travellers in an adaptive manner. Thus, short-term estimates should indicate the right order of magnitude and direction of the effect of counterfactual scenarios.

4.1 Defining the counterfactuals

We focus on the two extreme cases to find the upper and lower limits of the effects of cross-border traffic: We model what-would-have-been cases as two scenarios, where either all mobility is cancelled (zero mobility matrices) or where the mobility is restored to the 2019 pre-pandemic level. We compare these scenarios to a baseline scenario which is the model fitted to the real data from the year 2020 as described in Section 2.3.

We use the same sample of reproduction numbers, S R, sampled from the posterior distribution of the model fitted to the mobility (and health data) from year 2020 in all of our scenarios. For each of these samples R ∈ S R, we compute the number of infected individuals at each time step t given the reproduction number and mobility data I t , x ( R , D com , D long , i inflow , M • , 0 , • long ). Note that this computation is deterministic in our SIR model. We then estimate the posterior mean numbers of infected individuals, E ( I t , x ) ≈ 1 | S R | ∑ I t , x ( R , D com , D long , i inflow , M • , o , • long ) . We do all of our computations using | S R | = 1000 samples.

To construct a counterfactual model, we substitute the appropriate values of the mobility matrices D short, D long, i inflow, and M • , o , • long. To express the scenario for the border closure starting on day t, we modify the mobility data by filling the matrices corresponding to day t and later with zeros. For example, to model the effect of a border closure starting in May 2020, we set all values in the matrices starting from 1 May 2020 to zero. For the scenario of border reopening, the mobility matrices at time t and after are filled with data from 2019. We then recompute the expected number of infected individuals E(I) in Eq (20) with the modified matrices but the same sample of the reproduction numbers R.

We investigate the timing of the hypothetical interventions by varying the starting time at the first day of each month. We show the resulting counterfactual trajectories of I only for 50 days after the start of the scenario, assuming that after 50 days, ceteris paribus, assumptions for R become unrealistic.

 4.2 Results

For completely closing all border traffic see Fig 8, and returning back to the traffic patterns of 2019 see Fig 9 (see also S3 Appendix Fig J and K). In Finland and Norway, removing the border traffic in May and June leads to a large reduction in the number of infections during the fifty-day period in our model. Removing the border traffic in July still has a significant effect, but after August the effects are minimal. The reverse is true, but with slightly less dramatic changes, when we return back to 2019 border traffic. Here the increased travels during April have a larger effect than reducing the traffic would have had, presumably due to the real traffic already being low during April.

10.1371/journal.pcbi.1012182.g008 Comparison to counterfactual scenarios for the border closure.

Black line shows the mean number of infections E(It,x) in the baseline scenario. Colored lines presents the number of infections in counterfactual scenarios, during the 50 day interval starting with the implementation of restrictions. Dots mark the start of the restriction.

 10.1371/journal.pcbi.1012182.g009 Comparison to counterfactual scenarios for the border reopening.

Black line shows the mean number of infections E(It,x) in the baseline scenario. Colored lines presents the number of infections in counterfactual scenarios, during the 50 day interval starting with the implementation of border reopenings. Dots mark the start of the reopenings.

The results for Denmark are similar to those of Finland and Norway during April—October 2020, but with less dramatic changes. During the fall, Denmark would have slightly suffered from removing the border traffic according to our model. Sweden stands out from the other Nordic countries, because the border traffic has very little effect on its epidemic situation. Reduced border traffic in the fall, and increased border traffic during the spring and summer as this would have only slightly reduced the epidemic pressure inside the country.

 5 Summary and discussion

Due to the economic, legal and social complications related to any mobility restrictions it is crucial to understand the efficacy of such measures during pandemic spread. To this end, we have undertaken an in-depth modeling study of the effect of border closures in the four neighboring Nordic countries during 2020. We used two kinds of metrics to estimate the effect of the mobility: descriptive and counterfactual. Descriptive metrics, like the net mobility effect, focus on estimating the direct consequences, i.e., numbers of infections arriving or leaving the country. Such metrics can help decision makers to focus on the most important factors controlling the spread and growth of the infections. From our study we can conclude that for inter-country mobility to have an effect on the spread of the infections, three conditions have to be met: (1) a low number of new infections in the target country, (2) a high number of infections in the exporter country, and (3) a sufficiently high level of mobility between the countries to transfer infection.

An interesting detail that emerges from our study is that, in our model, commuters had a consistently low contribution to disease transmission. In our model each commuter spent only half a day in the destination country per border crossing, while long-term travellers spend their whole infectious period (8 days on average); this creates a ratio of 1 to 16, which is further amplified by the low number of identified commuters. This conclusion relies on assumptions about the behaviour of long-term travellers—in reality they could spend much less time in the destination country. There is however a lack of reliable data of this type on long-term travellers.

We should also mention that there is a multitude of possible secondary effects that we have not included in our modeling: arriving infections may cause an outbreak, the outbreak may cause tightening of the restrictions which in turn may lead to reduction of local transmission, which in turn can delay the herd immunity effects etc. While these secondary effects are often negligible in studies estimating the time until the outbreak starts, our focus on investigating the effect of inter-country mobility during the outbreak necessitates their consideration. To this end, we have included two counterfactual scenarios that should constitute bounds for the effects of mobility restrictions, at least in the medium-short time period.

Our rather simplified model does not include stratification by age, region, or detailed disease states. This may prevent our results from being quantitatively exact, but we believe that the qualitative properties would not be affected. It should be added that, during the research project, it was found that detailed data about inter-country mobility is scarce. For instance, the rate of vehicles crossing a border point may be known, but not the number of individuals, or the time they spend in the receiving country. More detailed data on such aspects could improve the modeling accuracy.

There are other important differences between the real epidemic and an SIR model. Super-spreading is suggested as a driving factor for the pandemic during the studied period [27]. Super-spreading implies a heavy tailed distribution of secondary infections, while the deterministic SIR model assumes a constant number. In addition, assortative mixing has been suggested as a factor influencing the infection dynamics [28]. In our model we use a time-varying reproduction number R t, which should be able to capture the fluctuations caused by super-spreading and assortative mixing. It should also aggregate the differences between countries, e.g. differences in climate and pandemic suppression policies.

Leaving one country to visit another may change the number, duration and context of social contacts. In our model, we assumed that people crossing the border have the same infectivity and susceptibility as the local people in the visited country (including those doing local travel). Collecting quantitative data on travellers’ behaviour abroad and estimating additional risks imposed by the travel itself would help in further planning of intervention measures.

National borders often constitute an ideal location for interventions due to a limited number of crossing points, and the existing infrastructure for controlling travel. Further, limiting movement within a country can be practically and legally more difficult than denying access to a country. The interventions can be achieved either via limiting the number of travellers or controlling and testing for symptomatic or asymptomatic passengers. However, assessing the effectiveness of inter-country travel restrictions can be more difficult than within-country travel, because one needs to combine data from multiple countries’ health officials. In addition, within-country mobility is much better studied than inter-country mobility. Despite these difficulties, we believe that our modelling study benefits both the assessment of intervention strategies at the borders and modelling epidemic spread within a given country by separating the effect of the external disease pressure and thus yielding more accurate reproduction numbers.

 Supporting information Mobility data and model.

(PDF)

 Derivation of the transmission model.

(PDF)

 Additional results.

(PDF)

 References European Commission, Directorate-General for Regional and Urban Policy, Peyrony, J., Rubio, J., Viaggi, R. The effects of COVID-19 induced border closures on cross-border regions—An empirical report covering the period March to June 2020. Publications Office; 2021. https://data.europa.eu/doi/10.2776/092793 Duchene S, Featherstone L, Freiesleben de Blasio B, Holmes EC, Bohlin J, Pettersson JHO. The impact of public health interventions in the Nordic countries during the first year of SARS-CoV-2 transmission and evolution. Eurosurveillance. 2021;26(44). doi: 10.2807/1560-7917.ES.2021.26.44.2001996 34738512 Duchene S, Featherstone L, Freiesleben de Blasio B, Holmes EC, Bohlin J, Pettersson JHO. Assessment of Coronavirus Disease 2019 Intervention Strategies in the Nordic Countries Using Genomic Epidemiology. Open Forum Infect Dis. 2021;9(3). doi: 10.1093/ofid/ofab665 McLaughlin A, Montoya V, Miller RL, Mordecai GJ, Consortium CCGNC, Worobey M, et al. Genomic epidemiology of the first two waves of SARS-CoV-2 in Canada. eLife. 2022;11:e73896. doi: 10.7554/eLife.73896 35916373 Nonnenmacher L, Baumann M, le Bihan E, Askenazy P, Chauvel L. Cross-border mobility in European countries: associations between cross-border worker status and health outcomes. BMC Public Health. 2021;21(1):588. doi: 10.1186/s12889-021-10564-8 33761912 Kucharski AJ, Jit M, Logan JG, Cotten M, Clifford S, Quilty BJ, et al. Travel measures in the SARS-CoV-2 variant era need clear objectives. The Lancet. 2022;399(10333):1367–1369. doi: 10.1016/S0140-6736(22)00366-X 35247312 Shiraef MA, Friesen P, Feddern L, Weiss MA, Janabi HA, Beling E, et al. Did border closures slow SARS-CoV-2? Scientific Reports. 2022;12(1):1709. doi: 10.1038/s41598-022-05482-7 35105912 Medeiros E et al. Covidfencing effects on cross-border deterritorialism: the case of Europe. European Planning Studies. 2021;29(5):962–982. doi: 10.1080/09654313.2020.1818185 Scalia Tomba G, Wallinga J. A simple explanation for the low impact of border control as a countermeasure to the spread of an infectious disease. Mathematical Biosciences. 2008;214(1):70–72. doi: 10.1016/j.mbs.2008.02.009 18387639 Russell TW, Wu JT, Clifford S, Edmunds WJ, Kucharski AJ, Jit M. Effect of internationally imported cases on internal spread of COVID-19: a mathematical modelling study. The Lancet Public Health. 2021;6(1):e12–e20. doi: 10.1016/S2468-2667(20)30263-2 33301722 Nordensvärd J, Sefton T, Godenhjelm S. Interpreting the state–citizen nexus in contemporary Nordic legal and social citizenship: the case of divergence in restriction on freedom of movement as a mitigation policy in the COVID-19 pandemic. Journal of International and Comparative Social Policy. 2023;39(1):28–41. doi: 10.1017/ics.2023.5 Heinikoski S, Hyttinen T. The Impact of Covid-19 on the Free Movement Regime in the North: Analysis of Border Closures in Denmark, Finland, Norway and Sweden. Nordic Journal of International Law. 2022;91(1):80–100. doi: 10.1163/15718107-91010004 Mishra S, Scott JA, Laydon DJ, Flaxman S, Gandy A, Mellan TA, et al. Comparing the responses of the UK, Sweden and Denmark to COVID-19 using counterfactual modelling. Scientific Reports. 2021;11(1):16342. doi: 10.1038/s41598-021-95699-9 34381102 Aagesen HW, Järv O, Gerber P. The effect of COVID-19 on cross-border mobilities of people and functional border regions: the Nordic case study from Twitter data. Geografiska Annaler: Series B, Human Geography. 2022; p. 1–23. Chang S, Vrabac1 D, Leskovec J, Ugander J. Estimating geographic spillover effects of COVID-19 policies from large-scale mobility networks. In: 37th AAAI Conference on Artificial Intelligence; 2023. Reichmuth ML, Hodcroft EB, Riou J, Neher RA, Hens N, Althaus CL. Impact of cross-border-associated cases on the SARS-CoV-2 epidemic in Switzerland during summer 2020 and 2021. Epidemics. 2022;41:100654. doi: 10.1016/j.epidem.2022.100654 36444785 Sattenspiel L, Dietz K. A structured epidemic model incorporating geographic mobility among regions. Mathematical Biosciences. 1995;128(1-2):71–91. doi: 10.1016/0025-5564(94)00068-B 7606146 Swedish covid-19 hospitalization data from Socialstyrelsen, dataset: Nyinskrivningar och avlidna under pandemin, https://www.socialstyrelsen.se/statistik-och-data/statistik/statistik-om-covid-19/, Norwegian covid-19 data from Folkehelseinstituttet (the Norwegian Institute of Public Health) https://github.com/folkehelseinstituttet/surveillance_data. Our World in Data. https://github.com/owid/covid-19-data/blob/master/public/data/owid-covid-data.csv. Vargas Bernal E, Saucedo O, Tien JH. Relating Eulerian and Lagrangian spatial models for vector-host disease dynamics through a fundamental matrix. Journal of Mathematical Biology. 2022;84(7):57. doi: 10.1007/s00285-022-01761-z 35676373 Molla J, Ponce de León Chávez A, Hiraoka T, Ala-Nissila T, Kivelä M, Leskelä L. Adaptive and optimized COVID-19 vaccination strategies across geographical regions and age groups. PLOS Computational Biology. 2022;18(4):1–19. doi: 10.1371/journal.pcbi.1009974 35389983 Vihola M. Robust adaptive Metropolis algorithm with coerced acceptance rate. Statistics and Computing. 2012;22(5):997–1008. doi: 10.1007/s11222-011-9269-5 Creswell R et al. Heterogeneity in the onwards transmission risk between local and imported cases affects practical estimates of the time-dependent reproduction number. Phil. Trans. R. Soc. A 380: 20210308. doi: 10.1098/rsta.2021.0308 35965464 Thompson RN et al. Improved inference of time-varying reproduction numbers during infectious disease outbreaks. Epidemics 29: 100356. doi: 10.1016/j.epidem.2019.100356 31624039 Britton T, Leskelä L. Optimal Intervention Strategies for Minimizing Total Incidence During an Epidemic. SIAM Journal on Applied Mathematics. 2023;83(2):354–373. doi: 10.1137/22M1504433 Wong F, Collins JJ. Evidence that coronavirus superspreading is fat-tailed. Proc Natl Acad Sci U S A. 2020 Nov 24;117(47):29416–29418. doi: 10.1073/pnas.2018490117 33139561 Britton T, Ball F, Trapman P. A mathematical model reveals the influence of population heterogeneity on herd immunity to SARS-CoV-2 Science 369, 846–849 (2020). doi: 10.1126/science.abc6810 32576668 10.1371/journal.pcbi.1012182.r001 Decision Letter 0 Pitzer Virginia E. Section Editor Struchiner Claudio José Academic Editor 2024 Pitzer, Struchiner This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Submission Version 0

6 Feb 2024

Dear Dr. Shubin,

Thank you very much for submitting your manuscript "The influence of cross-border mobility on the COVID-19 epidemic in Nordic countries" for consideration at PLOS Computational Biology.

As with all papers reviewed by the journal, your manuscript was reviewed by members of the editorial board and by several independent reviewers. In light of the reviews (below this email), we would like to invite the resubmission of a significantly-revised version that takes into account the reviewers' comments.

All reviews emphasize the need for a clear and thorough discussion of the work's limitations. This discussion must assess the robustness of the conclusions in light of those limitations, especially regarding their consequences for decision-making.

We cannot make any decision about publication until we have seen the revised manuscript and your response to the reviewers' comments. Your revised manuscript is also likely to be sent to reviewers for further evaluation.

When you are ready to resubmit, please upload the following:

[1] A letter containing a detailed list of your responses to the review comments and a description of the changes you have made in the manuscript. Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out.

[2] Two versions of the revised manuscript: one with either highlights or tracked changes denoting where the text has been changed; the other a clean version (uploaded as the manuscript file).

Important additional instructions are given below your reviewer comments.

Please prepare and submit your revised manuscript within 60 days. If you anticipate any delay, please let us know the expected resubmission date by replying to this email. Please note that revised manuscripts received after the 60-day due date may require evaluation and peer review similar to newly submitted manuscripts.

Thank you again for your submission. We hope that our editorial process has been constructive so far, and we welcome your feedback at any time. Please don't hesitate to contact us if you have any questions or comments.

Sincerely,

Claudio José Struchiner, M.D., Sc.D.

Academic Editor

PLOS Computational Biology

Virginia Pitzer

Section Editor

PLOS Computational Biology

***********************

All reviews emphasize the need for a clear and thorough discussion of the work's limitations. This discussion must assess the robustness of the conclusions in light of those limitations, especially regarding their consequences for decision-making.

Reviewer's Responses to Questions

Comments to the Authors:

Please note here if the review is uploaded as an attachment.

Reviewer #1: The manuscript: “The influence of cross-border mobility on the COVID-19

epidemic in Nordic countries” seeks to model the impact of between country border crossings on the spread of COVID-19 during the first year of the SARS-CoV-2 pandemic in 2020.

The authors use a rich set of border-crossing data available from four of the Nordic countries (Denmark, Norway, Sweden and Finland) to inform SIR models. These models are subsequently used to explore counterfactual scenarios relating to effects on the pandemic as a consequence of open or closed borders. Their findings suggests that border-crossing policies can have considerable influence on the evolution of a pandemic, more importantly, they find that there are certain instances where the influence of border-crossings on an epidemic are negligible. These findings are important for policy-makers and modelers alike to take into account when the next pandemic strikes. Moreover, there are only a few studies examining the effects of border crossings on pandemics such as COVID-19. Thanks to their cultural and geographic similarity, the Nordic countries provide excellent case studies that can potentially tease apart effects that can be relevant for many other western countries.

In my opinion, this is a very timely and interesting study examining cross-border effects on outbreaks, something that is lacking from the Nordic countries and therefore of great interest. By choosing four similar countries in terms of governance, geography and, to some extent, demography it is possible to control for the impact of country-specific measures on outbreaks. The manuscript is very well written and the model is thoroughly explained and appears to be sound. While there are limits as to what can go into a mathematical model of an infectious disease pandemic, there are some points I hope the authors could clarify and discuss to a greater extent than what has currently been done.

***Major revisions***

During the first year of the COVID-19 pandemic, there were many indications that super-spreading events were to some extent driving the pandemic (see 33139561[pmid]). How is this reflected by the model (if at all) and what impact can super-spreading events have on travel between countries? (i.e. can travel by boats, trains or planes function as super-spreading events? (see 35779143[pmid], 32726405[pmid]) How can such super-spreading events go on to drive secondary infections in the host countries? (This is also discussed somewhat in the referenced Creswell et al. article). It is not expected that the authors redo their analyses of course but a couple of sentences describing super-spreading and how this might be challenging to account for with standard SIR models would be helpful.

The ability to identify the true effect of boarder crossings on infectiousness is likely very difficult to assess as it is clearly challenging to separate local infections from imported. The values presented are therefore only projections assuming nothing but the reproductive number from the source country it seems. This should be highlighted as a premise for the study and reflected upon when the consequences and conclusions are presented, particularly for the counterfactual part. Traveling itself may introduce unidentified risks not accounted for and may not have been controlled for in the study. This should be pointed out in the discussion and recommendations section.

All countries considered are well-governed and although similar there are also some marked differences. While Finland and Norway are fairly similar in terms of both population size/density and geography (both countries also share borders with Russia), Sweden has a population almost double the size and thus lower population density. Denmark, on the other hand, is located on the European continent, away from Norway and Finland, and has an island-like structure with a markedly higher population density. As an EU member (as both Sweden and Finland also are), Denmark shares open borders with Germany in addition to Sweden. I miss more discussion on how these differences can impact the spread of COVID-19 internally in each country as a consequence of importations (See for instance 35229003[pmid]. It doesn’t have to be much but it should be mentioned how these differences are reflected in their model. The model is the same for all countries but the countries inform the model differently from the available data, can something have been lost?

What effect on the population does the length of immunity and the reproduction number in a country have on imported cases and their role in driving (or reducing) the pandemic? That is, if a fraction of the population is immune and immunity lasts for a long time (say 6-12 months) how does this influence the effects of border crossings on the evolution of the pandemic within a country? How can this differ from say shorter duration of immunity as witnessed by the more recent SARS-CoV-2 Omega strains? What about a country with a higher population density like Denmark, how might imported infections spread there compared to countries with a lower population density like Norway and/or Finland? Is it at all possible to separate the effects from importation in a country like Sweden with high infection rates from a country like Denmark with higher density, or will these countries appear the same from the proposed model?

Not much is revealed regarding the impact of Russian cross-border traffic in Norway and Finland in the main text, although some results are presented in the S1 appendix. How could these border crossings affect spread of COVID-19 and is it expected to differ from within Nordic countries border crossings?

***Minor revisions***

Is it Shubin Mikhail or Mikhail Shubin (see S1-S3 appendix authors).

In S1 appendix, 2. page 2. paragraph: make sure you indicate the “o” index in D_{t,d ←o} ^{long/short}

In S1 appendix, 2. last paragraph page 2: The superscripts ? (plural)

In S1 appendix, first paragraph, page 7: Note that the because…

It is not always completely apparent what the figures in S3 appendix designate, some more information in the figure legends would be appreciated.

***Remarks that the authors can decide whether have merit***

The initial reproduction number is defined to be Normal with mean=2 and sd=0.5, this can some times be negative, which is nonsensical, why not use log-normal?

In equation 13 the letters of the different terms end up with the same letter “I” that also designates infectiousness.

Reviewer #2: This is an interesting paper that uses a tractable model to investigate the impact of public health interventions, people movement in particular, on the spread of SARS-CoV2 in the Nordic countries. I find this work interesting and relevant. The results seem sensible, but they are very difficult to evaluate in absence of some key details that I list below, and which I consider essential before I am able to recommend revisions.

- Section 3.1.2 requires a more detailed explanation of how M is estimated. I suggest spelling out each term in equations 16 and 17.

- Similar to above, the final sentence in section 3.1.3, ‘Prevalence of infection among long-term travellers from outside Nordic countries is provided in the data’, is not entirely clear. Is this part of the data that was fed to the model and is it thus fixed? This is not the impression I got from reading the rest of the methods, but I could be entirely wrong.

- I appreciate the graphical depiction of the model (Fig 2). However, it would be very helpful to have an expression for what the likelihood function is here. There are many parameters and if the data are just the number of hospitalisations per day, then stating out the likelihood here would help put everything into context.

- In Fig 2 I can see that there are several fixed components in the model. Can the authors comment on their uncertainty? For example, whether the hospitalisation probability or its probability after t days of infection have associated uncertainties. In that case, what would be the effect of treating these as stochastic nodes? Presumably not much, but I think this point would help the reader understand the extent to which the model is realistic.

- The authors have commented on variations of their model due to age and regional stratification. However, there are other important factors, like those that break the SIR assumptions. I suggest adding a few sentences on whether these can have an impact here.

- There is substantial evidence that Sweden had the highest viral genetic diversity of the virus among Nordic countries in the early stages of the pandemic. This indicates that it acted as a net exporter of the virus. Of course, these dynamics probably changed rapidly, with local transmission playing an increasingly big role in each jurisdiction. In the light of this work, this can simply mean that the number of importations was not as important as the fact that they actually resulted in ongoing transmission in each country. Can the authors comment on how the genetic evidence stacks up and its implications?

- There are a few minor typological or grammatical errors that can be easily fixed with careful proofreading. E.g. line 243 ‘between by country’ – remove the ‘by’.

Reviewer #3: As an epidemiologist reviewer, my primary focus is on the results and reporting. The paper titled 'The Influence of Cross-Border Mobility on the COVID-19 Epidemic in Nordic Countries,' authored by Dr. Mikhail Shubin and colleagues, presents an elegant study. The authors have quantitatively estimated the impact of cross-border traffic on the number of COVID-19 infections in Nordic countries during the first year of the pandemic. The study features a clear and concise presentation of the conditions necessary for inter-country mobility to influence the spread of infections. The listed criteria, including (i) very low new infections in the target country, (ii) a significant disparity in infections between the source and target countries, and (iii) a substantial level of mobility, provide a robust framework for understanding when inter-country mobility becomes a critical factor in infection transmission. The authors also emphasize the increase in imported infections in Nordic countries during the summer of 2020, attributed to infected travelers from Sweden and, later in 2022, from Denmark. This information is concisely described and offers valuable insights into the source of these infections.

Furthermore, the authors present a particularly noteworthy scenario in which Meff > 1 and Reff < 1, indicating that the epidemic could propagate solely through inter-country mobility. This specific situation was observed in Finland and Norway during the summer of 2020. Additionally, substantial differences in prevalence between local and traveling populations are noted, with Nordic travelers being more likely to be infectious in Norway and Finland. In contrast, the opposite effect is observed in Sweden. The authors also explore counterfactual scenarios for border closures and reopenings.

In summary, this article makes a valuable contribution to our understanding of the impact of international mobility on the COVID-19 pandemic, offering a comprehensive analysis of the specific case of the Nordic countries. The manuscript is suitable for publication in PLOS Computational Biology.

Suggestions for improvement:

1. One aspect that the article does not address is the distinction between different COVID-19 virus variants. While the dominance of the Wuhan variants is evident for most of 2020, the emergence of the VOC Alpha variant in December is not sufficiently highlighted.

2. Furthermore, enhancing the work by including comparisons to other models or data from different countries could enrich the discussion and conclusions.

Reviewer #4: Shubin et al. explore the effects of cross-border movement on the trajectory of the early SARS-CoV-2 pandemic in Nordic countries. They employ data-driven modelling of concurrent infection trajectories in each country, including parameters for short and long term travel between each and the rest of the world. They also consider counterfactual scenarios to corroborate their results. Overall, they show that the effects of travel can be significant, but only if particular conditions are met. These results add nuance to the broader discussion on border restrictions during the pandemic.

I want to commend the authors for undertaking this work given the difficulty in collating the data here. I would like to see it published following revision. While the science appears sound, the manuscript is lacking detail or unclear in places. My comments largely pertain to making it clearer.

Major Comments:

L115-121: The text around Gaussian mixture models here is quite confusing. It would b

 10.1371/journal.pcbi.1012182.r002 Author response to Decision Letter 0 Submission Version 1

15 Apr 2024

Submitted filename: replies_with_appedix.pdf

 10.1371/journal.pcbi.1012182.r003 Decision Letter 1 Pitzer Virginia E. Section Editor Struchiner Claudio José Academic Editor 2024 Pitzer, Struchiner This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Submission Version 1

20 May 2024

Dear Dr. Shubin,

We are pleased to inform you that your manuscript 'The influence of cross-border mobility on the COVID-19 epidemic in Nordic countries' has been provisionally accepted for publication in PLOS Computational Biology.

Before your manuscript can be formally accepted you will need to complete some formatting changes, which you will receive in a follow up email. A member of our team will be in touch with a set of requests.

Also note that Reviewer 4 has some minor suggestions that can be addressed at the proof stage prior to publication.

Please note that your manuscript will not be scheduled for publication until you have made the required changes, so a swift response is appreciated.

IMPORTANT: The editorial review process is now complete. PLOS will only permit corrections to spelling, formatting or significant scientific errors from this point onwards. Requests for major changes, or any which affect the scientific understanding of your work, will cause delays to the publication date of your manuscript.

Should you, your institution's press office or the journal office choose to press release your paper, you will automatically be opted out of early publication. We ask that you notify us now if you or your institution is planning to press release the article. All press must be co-ordinated with PLOS.

Thank you again for supporting Open Access publishing; we are looking forward to publishing your work in PLOS Computational Biology. 

Best regards,

Claudio José Struchiner

Academic Editor

PLOS Computational Biology

Virginia Pitzer

Section Editor

PLOS Computational Biology

***********************************************************

Please note that Reviewer 4 has some minor suggestions that can be addressed at the proof stage prior to publication.

Reviewer's Responses to Questions

Comments to the Authors:

Please note here if the review is uploaded as an attachment.

Reviewer #1: My comments have been answered.

Reviewer #2: I would like to thank the authors for their effort in improving the clarity. They have addressed many technical comments from four reviewers, which implied substantial editing and careful responses.

I am satisfied with the work they have done, and particularly their explanation of parameter identifiability, the definition of the likelihood functions and the potential impact of violations to this SIR model (e.g. superspreading). I do not have additional comments and am happy to recommend this paper for publication.

Reviewer #3: The authors meticulously addressed all my concerns. After careful consideration, I find no further points to critique. I commend the authors for their diligent efforts and thoroughness of work. Given the quality and significance of their findings, their manuscript deserves publication in the PLOS CB journal. Congratulations to the authors on their excellent contribution to the field.

Reviewer #4: The authors have addressed all of my comments. I think the manuscript has improved and I do not need to see it again.

It will be nice to see an epidemiological study looking at the Nordic countries in addition to the genomic work in the literature, and this paper unifies the results well.

I only have a couple of minor corrections to make, and urge the authors to check carefully for typos at the proofing stage, since the expression is sloppy in parts.

Minor:

L5: economical

L96: Stick to SARS-CoV-2 as this is the virus, like you have listed for other viruses like Ebola, SARS, and so on. Covid-19 is the name of the corresponding disease. I am sorry to be pedantic, but this is an important distinction to be consistent with the rest of the literature.

L104: Millions → million

L281: “exporter of mobility” doesn’t make sense. Maybe ‘exporter of infections’?

Fig3: I would suggest a line pointing out that negative values correspond to the export of infections. This would make things much clearer, since I found I needed to reference the figures before parsing the text.

L415: “Benefited” -> “exported”, as per my comment in initial reviews.

**********

Have the authors made all data and (if applicable) computational code underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data and code underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data and code should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data or code —e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: Yes

Reviewer #3: Yes

Reviewer #4: Yes

**********

PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: No

Reviewer #3: No

Reviewer #4: No

 10.1371/journal.pcbi.1012182.r004 Acceptance letter Pitzer Virginia E. Section Editor Struchiner Claudio José Academic Editor 2024 Pitzer, Struchiner This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

10 Jun 2024

PCOMPBIOL-D-23-01783R1

The influence of cross-border mobility on the COVID-19 epidemic in Nordic countries

Dear Dr Shubin,

I am pleased to inform you that your manuscript has been formally accepted for publication in PLOS Computational Biology. Your manuscript is now with our production department and you will be notified of the publication date in due course.

The corresponding author will soon be receiving a typeset proof for review, to ensure errors have not been introduced during production. Please review the PDF proof of your manuscript carefully, as this is the last chance to correct any errors. Please note that major changes, or those which affect the scientific understanding of the work, will likely cause delays to the publication date of your manuscript.

Soon after your final files are uploaded, unless you have opted out, the early version of your manuscript will be published online. The date of the early version will be your article's publication date. The final article will be published to the same URL, and all versions of the paper will be accessible to readers.

Thank you again for supporting PLOS Computational Biology and open-access publishing. We are looking forward to publishing your work!

With kind regards,

Zsofia Freund

PLOS Computational Biology | Carlyle House, Carlyle Road, Cambridge CB4 3DN | United Kingdom ploscompbiol@plos.org | Phone +44 (0) 1223-442824 | ploscompbiol.org | @PLOSCompBiol