Four million COVID-19 cases in four European cross-border regions reveal minimal cross-border transmission effect on domestic burden: space-time cluster analysis

Abstract

Introduction

Restrictions of cross-border mobility were used during the COVID-19 pandemic to slow down transmission between countries. Such restrictions came with significant societal costs, and their effect on in-country burdens of COVID-19 has been questioned. Our study aims to assess whether cross-border transmission contributed significantly to in-country burdens of COVID-19 in four large Euroregions shared by the Netherlands, Germany and Belgium.

Method

Geographical information systems (GIS) and Besag-York-Mollie (BYM) models were used to visualize the spatial distribution of COVID-19 cases from January 2020 through October 2022. Space-time cluster analysis with SaTScan was used to identify significant clusters of COVID-19 cases and determine whether they included cross-border areas. Over 4 million cases drawn from the national infectious disease registries were analysed for this study.

Results

Incidence of COVID-19 increased and decreased independently within each country, except for one simultaneous peak in January 2022 due to the Omicron variant. Eight significant space-time clusters were identified, four of which were cross-border clusters. All eight clusters occurred during periods of high in-country transmission.

Discussion

Our findings showed that cross-border transmission did not contribute in significant ways to in-country burdens of COVID-19 burden in the regions studied. The identified clusters corresponded to in-country waves of infection. Our results challenge the assumption that border closures were crucial in controlling the pandemic and highlight the need for coordinated public health responses that balance the effect of interventions with societal costs.

Introduction

About 35% of Europe’s citizens live inside an area defined by the European Union (EU) as having a land border. Within the EU, about one-third of its regions are considered border regions by the EU [1, 2]. Arguments for establishing the EU were rooted in economic advantages, social and cultural integration, political stability resulting in freedom of movement across borders for people, goods, and services under the framework of the Schengen agreement.

However, living in a border region may also pose extra challenges and health-related issues arising from a confluence of epidemiological, socioeconomic, infrastructural and cultural factors. This became even more clear during the COVID-19 pandemic years 2020 and 2021, when European countries implemented a variety of measures to limit border-crossing, motivated by the idea of controlling the spread of the virus. Countries installed restrictive measures like border closures, travel restrictions, quarantine requirements and health screenings. The border closure measures had a major impact on the socio-economic and cultural mobility of citizens, limiting their ability to commute, visit and take care of friends or family, or cross borders to see their family doctor [3–5].

Border restrictions, despite their obvious trade-offs, may have been seen as a legitimate way to mitigate the transmission of SARS-CoV-2, as studies showed that traveling between (EU) countries played a role in the spread of SARS-CoV-2 across Europe [6, 7]. The primary argument was to contain the virus by reducing cross-border movement, thereby limiting the potential for imported cases, preventing new outbreaks and especially preventing the importation of SARS-CoV-2 variants and thus protecting the public health systems from becoming overwhelmed. Restrictions on cross-border mobility were often seen as a ‘natural’ extension of interventions on a national level. Evidence shows, for example, that Dutch holiday-goers who had been to Austria or Italy during the early weeks of the COVID-19 pandemic became infected with SARS-CoV-2 at their holiday destinations and introduced the virus into their home country upon returning home [7], although the relative contribution to the total burden of transmission remains unclear as does effectiveness of border closures in mitigating this unspecified risk. A Dutch cross-border study based on SARS-CoV-2 serology and questionnaire outcomes from almost 10,000 citizens showed no evidence for ‘cross-border mobility’ as a significant or substantial exposure factor for SARS-CoV2 transmission [8].

A study on the societal impact of cross-border restrictions in the border region of the Meuse-Rhine Euroregion (Netherlands, Germany, Belgium) and the effect of cross-border restrictions during the COVID-19 pandemic confirmed that citizens perceived these cross-border interventions as impactful and disruptive to their everyday wellbeing and presented citizens with financial consequences and logistical challenges, sometimes affecting access to health care and education [9]. While policy and legislation often focus on geopolitical borders, the authors of the said studies argue that effective policy must take into account the fluid nature of borders in citizens’ daily lives.

The relevance of these issues is underlined by the stance the EU takes in their Regulation on serious cross-border threats, which states that uncoordinated public health crisis management measures taken by individual EU member states may affect other member states. Coordination at EU level should ensure that member state measures are proportionate and limited to public health risks resulting from serious cross-border threats. This should not conflict with EU obligations and rights, such as the free movement of persons, goods and services [10]. The EU itself has no mandate to decide on border closures or travel restrictions, as such decisions rest with national governments. Since such decisions often directly affect border regions, regional and cross-border perspectives warrant consideration in the design of national measures.

To deliver evidence in support of these EU objectives, we explored the scale and scope (as well as associated patterns) of the cross-border transmission of SARS-CoV-2 between the Netherlands, Germany and Belgium in excess of the probability level of domestic and/or accidental transmission. Our research question therefore was whether there is evidence of a significant and substantial role of cross-border transmission of SARS-CoV-2 in relation to the burden of COVID-19 within a country, in order to inform future policy-makers regarding the consideration and justification of border measures.

Methodology

Data

To answer our research question, we analysed and visualised spatial distribution patterns of COVID-19 cases using a Geographic Information System (GIS) described in more detail below. The geographic area covered by our analyses are the four cross-border regions, the so-called Euroregions, between the Netherlands, Germany, and Belgium. These are the Meuse-Rhine Euroregion, the Rhine-Meuse North Euroregion, the Rhine-Waal Euroregion and the EUREGIO Euroregion (Münster (DE) – Twente (NL)Region). The border regions of these four Euroregions are home to approximately 10.1 million residents, 3.3 million of whom live in the Netherlands, 4.8 million in Germany and 2 million in Belgium. These four regions were chosen as they share common borders in close proximity to each other and were geographically distant from the borders with other countries like France or Luxembourg. Mobility across those more distant borders was likely limited, particularly during major phases of the pandemic. Numbers of COVID-19-positive cases were based on publicly available data of notifications to the national infectious disease registers in the time period between 2^nd January 2020 and 2^nd October 2022 (RIVM in the Netherlands, Robert Koch Institute in Germany and Sciensano in Belgium). During the over two-and-a-half year study period – which was from 1^st of March 2020 (start of the pandemic) until 2^nd October, 2022, in total 4,095,607 COVID-19 cases were included. Of this total number of cases, 1,548,412 were from the Netherlands, 1,780,009 from Germany and 767,186 from Belgium. Data was available at the level of municipalities in Belgium and the Netherlands, and at the level of counties in Germany.

Animated map

To visually display the dynamics of COVID-19, we calculated the 7-day incidence for the study period between 1^st March 2020 and 2^nd October 2022. To adjust for the uneven distribution of populations between the relatively small Dutch and Belgian municipalities versus the larger German counties, we used the Besag-York-Mollie (BYM) model [11]. The BYM model, which adjusts for structured and unstructured effects, weights the incidence of one area to the average incidence of the neighbouring areas (structured effects) and also shrinks the incidence towards the global mean (unstructured effects) [12]. This approach has been used extensively, e.g., for small-area disease-mapping of test-positivity for COVID-19 in New York [13], COVID-19 hospitalisations in north-eastern Germany [14], as well as for a wide range of other diseases such as cancer [15] and cardiovascular mortality [16].

Specifically, the BYM model used for our analysis is a Poisson model, consisting of the number of positive COVID-19 cases per municipality or county and the number of expected cases within the last 7 days. Neighbouring areas were defined as areas sharing a common edge or border [12]. In total, 946 models were calculated in the R-INLA package [17]. The incidence rates were displayed in in the ggplot2 package [18] in RStudio (Version: 2023.12.1 + 402). The animated GIF map was then created using the gganimate package [19].

Static map

In addition to the animated map, we created a static map for six different time periods covering the period from March 2020 to September 2022, to visually display the overall temporal dynamics of the pandemic. The methodology was the same as for the animated map with the difference that we calculated the incidence for the respective time periods and not the 7-day incidence. The calculation was performed again using the R-INLA package [17]. The incidence rates were displayed in in the ggplot2 package [18] in RStudio (Version: 2023.12.1 + 402).

Space-time cluster analysis

The space-time scan statistic was used to detect high-risk areas where COVID-19 cases clustered in space and time, indicating if and when cross-border spread of SARS-CoV-2 was likely to occur [20, 21]. We used a retrospective space-time Poisson model, consisting of the number of COVID-19 cases per day and area, the number of inhabitants per area and the centroid coordinates for each municipality or county [22]. We evaluated different possible cluster sizes. We found that the standard setting, i.e., of including up to 50% of the population at risk and up to 50% of the time period in a cluster, delivered results of no practical use since it detected only one giant cluster. Given the geography of our study area and the relatively small geographical size of the Dutch and Belgian municipalities compared to the larger German counties, we found that restricting the maximum cluster radius to 30 km with no temporal restrictions delivered reasonable results. Alternative SaTScan settings were also tested, such as a scanning window of 20 km radius and a maximum temporal window of 7, 14 and 31 days. However, this tended to deliver many very small clusters of only a handful of municipalities with very few cases inside, which are of very little practical use. We therefore chose to use a 30 km radius to align with the buffer zone proposed by the European Commission and applied by Euregio programmes to define border areas in EU cross-border cooperation (e.g. Interreg A) [23], and opted not to apply any temporal restriction. As there is, unlike in regression analysis, no metric available such as Aikaikes Information Criterion (AIC) to evaluate the best possible set of combinations for the data, we chose to use a setting, which can be justified by the EUs cross border policy to align our analysis with this policy. For the calculation of the space-time clusters, we used the SaTScan software (Version 10.1.3). The results of the cluster analysis were imported into RStudio and were visualised with the ggplot2 package.

Temporal analysis of COVID-19 incidence

To visualise how the 7-day incidence increased over time in the three countries of the border region, we created a line graph for the areas of the Netherlands, Belgium, and Germany and the EU-average. The country-specific 7-day incidence was calculated for each day. The 7-day incidence for the EU average was downloaded from Our World in Data.

The resulting line graph was created using the ggplot2 package in R. Additionally, we added a cylinder to display the time where space-time clusters occurred. During the COVID-19 pandemic, border control policies in the EU remained under the authority of the individual member states, as the EU has no legal mandate to impose uniform border closures. This led to differences in national responses: Belgium formally implemented full border closures for non-essential travels during two periods (March-June 2020 and January-April 2021), though exemptions were allowed for essential reasons. Germany introduced temporary border checks and entry restrictions, including testing requirements, without officially closing its borders. The Netherlands maintained open borders while issuing strong travel advisories. These national measures provide additional context when interpreting the cross-border dynamics presented in fig. 4.

Fig. 4.

Temporal distribution of COVID-19 for the three countries within the four Euroregions study area and the EU. The black cylinder shows the occurrence of all significant space-time clusters. The dark red cylinder shows the closing of the Belgian border

Results

Animated map of COVID-19

The animated map - which is included as Fig. 1 - shows that an increasing incidence in autumn of 2020 was first observed in the southern part of the Belgian side and with some delay in the Dutch region. In the winter period of 2020/2021 until spring of 2021, a higher incidence was observed scattered across the Belgian and Dutch sides, whereas incidence in Germany remained relatively low. During the summer period of 2021, incidence across the study area remained very low. In the autumn of 2021, incidence increased again on the Belgian and Dutch sides but remained low in Germany. Towards the end of January 2022, incidence increased to an all-time high across the entire study area. While incidence decreased relatively sharply on the Belgian and Dutch sides towards April 2022, with some municipalities in the Belgian and Dutch sides still displaying very local and short-lived increases, incidence generally remained higher on the German side, with a clear difference between the border of Germany and that of the Netherlands. The animated map clearly shows that incidence tended to increase and decrease independently in each country, except in January 2022, when incidence increased in all parts of the study area.

Fig. 1.

Animation COVID-19 incidence within the four euroregions study area of Netherlands, Germany, and Belgium. Time period 01–03-2020 until 30–09-2022

Static map

The static map for the six different time periods (Fig. 2) shows that the starting period of March – June 2020 was generally characterized by low incidences of reported cases. In the period of July – December 2020, a higher incidence could be observed in the southern part of Belgium while incidence remained low in Germany. Between January – June 2021, incidence in Germany remained low, but was slightly higher in the Netherlands and Belgium. A sharp contrast could be observed between July – December 2021, with higher and relatively similar incidences in the Netherlands and Belgium and lower incidence in Germany. Between January – June 2022, incidences increased across all of the study area, with the highest incidences being observed in the Netherlands and parts of Germany, while Belgium showed comparably lower incidence. In the last period, July – September 2022, incidences decreased again across the entire study area, although it should be noted that incidence was higher in Germany compared to the Netherlands and Belgium.

Fig. 2.

COVID-19 incidence for six different time periods within the four euroregions study area of Netherlands, Germany, and Belgium. Time period 01–03-2020 until 30–09-2022

Space-time clusters of COVID-19

In total, we identified 8 statistically significant (p < 0.05) space-time clusters (Fig. 3). All clusters lasted on average three to four months. Only one cluster started in November 2021 (only within the country of Belgium and not cross-border), while the remaining seven clusters all started in January 2022 when transmission within all four Euroregions was very high because of the ‘Omicron variant’ of Sars-CoV-2 virus. No clusters were detected for the longer time period of over one-and-a-half years between January 2020 and October 2021. Four of the clusters included a cross-border area; two between Germany and the Netherlands, one including parts of Germany, the Netherlands and Belgium, and one including areas in Germany and one municipality in Belgium. Four clusters did not include cross-border areas.

Fig. 3.

Space-time clusters of COVID-19 within the four Euroregions study area of Netherlands, Germany, and Belgium. Clusters are sorted according to their log-likelihood ratio

Temporal analysis of COVID-19 incidence

All three countries displayed a similar temporal pattern of occurrence of COVID-19 cases, with regular incidences before the end of 2022, followed by a steep increase towards the end of 2021. The surge in COVID-19 cases and high prevalence in many EU countries at the end of 2021 and the first half of 2022 can be attributed to the arrival of the Omicron variant. The surge in COVID-19 cases and high prevalence in many EU countries at the end of 2021 and the first half of 2022 can be attributed to the arrival of the Omicron variant. Only Belgium saw a spike in the autumn of 2021, which was independent of the other countries (Fig. 4). Germany also saw a smaller wave of rising incidences in the second quarter of 2022 which was more pronounced than that in Belgium and the Netherlands.

Discussion

In this paper, we used space-time analysis to identify clusters in over 4 million COVID-19 cases in four Euroregions of the Netherlands, Germany, and Belgium during a study period of more than two-and-a-half years. Specifically, we explored whether the spatial patterns we found were suggestive of a significant contribution of cross-border transmission of SARS-CoV-2 to in-country burdens of COVID-19.

Our analysis detected a total of eight spatial clusters, all of which occurred in the same time period, with in-country incidences within clusters largely following the same temporal pattern, while no clusters were seen over roughly the first period – lasting more than one and half years – of the COVID-19 pandemic. Clusters within countries and across borders occurred during the peak of the pandemic in late 2021 and the first half of 2022 as was the case in many other countries. The absence of space-time clusters before November 2021—a period that coincided with stricter border measures, particularly in Belgium—may indicate a potential mitigating effect of such policies. However, as our study was not designed to establish causality, this observation should be interpreted with caution. Other factors, including overall case levels, national lockdowns, testing regimes, and reduced mobility, may also have contributed to the delayed emergence of clusters. Notably, no clusters were observed in areas both with and without border measures.

This surge can be attributed to the arrival of the Omicron variant of SARS-Cov2. Omicron’s high transmissibility, led to a rapid spread within populations globally [24]. The surge in cases can also be attributed to waning immunity from previous infections or vaccinations [25], as well as challenges in implementing effective public health measures [26]. In addition, factors such as increased travel, relaxed restrictions, and pandemic fatigue further contributed to the widespread transmission of the virus during this period [27–29].

To satisfy the hypothesis that cross-border transmission contributed in significant ways, we would have expected space-time clusters to shift in time and space, with a clear direction from one in-country sub-area to another within a given cluster. However, given that incidence only spiked simultaneously in different national sub-areas once in the same period, that is, at a time when Omicron surged, our analysis provides no evidence for significant and substantial contribution of cross-border transmission to in-country burden of COVID-19 burdens within each country. Rather, it is in line with the assumption that by that time SARS-CoV-2 had become ubiquitous. This is highlighted by the animated map, which is void of incidence gradients shifting across borders, suggesting that, in the time periods before the end of 2021, higher incidences tended to occur relatively independently of border areas and incidences seemed more driven by in-country interventions and virus dynamics rather than by cross-border transmission. Even in the winter period stretching from 2021 into 2022, no differential involvement of national subareas could be observed within clusters. Although we did not include phylogenetic analyses to trace viral transmission routes, as comparable cross-border genomic sequencing data were not available during the study period, the emergence of viral genotypes followed distinct waves that were largely consistent across the study area and Europe, with temporal delays generally limited to one or two weeks. The observed spatiotemporal patterns do not indicate a major role of cross-border transmission in shaping the national COVID-19 burden, may suggest that border closures were of limited effectiveness in this cross-border context. This is consistent with a study using qualitative data and geographical visualisations of day-to-day infection rates data [5].

Our study has several limitations. First, data gathering took place at different geographical levels (municipality versus district level). Spatial resolution was higher for the Netherlands and for Belgium, where data were available at a municipality level, whereas German data were only available at a county level. However, while spatial resolution is lower for German counties, which usually comprise several municipalities, the impact on our findings was likely limited, given the high number of cases in the analysis. Second, subtle national differences in official case definitions, testing regimes, or testing compliance may have contributed to some degree of bias potentially compromising the full comparability of notification data between countries. Third, both a GIS study [30] and a longitudinal study [9] showed that cross-border mobility decreased significantly during some periods of the COVID-19 pandemic. Nevertheless, some degree of cross-border mobility always persisted, such that our results remain valid. Fourth, the circular scanning window of the space-time scan statistic may, by definition, introduce bias, since independent outbreaks in opposite (national) sub-areas may be subsumed under the same geographical cluster. Clusters enclosing different national sub-areas therefore do not necessarily reflect cross-border spread. For example, we cannot rule out the possibility that transnational cluster 3 represents two different national outbreaks, one in the (far) eastern German part of the cluster and one in the (far) western part in Belgium. As both outbreaks are subsumed under the same space-time cluster, they may have been mistakenly interpreted as a common cross-border event, even though this may not be the case.

Fifth, data was only available in an aggregated form, and not on an individual level. To be able to clearly trace and identify transmission routes between cases and their contacts, it would be necessary to have detailed personal information, including residential addresses of cases and their contacts.

Sixth, we only had access to small-area data for our study region, but not for the entire countries of the Netherlands, Belgium and Germany. Therefore, we could not analyse to what extent the spatial and temporal dynamics observed in our study region were representative of countries as a whole. This limitation may affect the detection of clusters, as SaTScan is sensitive to the number of observations included in the analysis. Consequently, we could not verify whether we would detect the same clusters if we had include data for the entire countries. Finaly, epidemic dynamics outside the study regions may have influenced the observed spatial patterns to some extent, although a significant external influence is considered unlikely given the broadly synchronous epidemic waves across the three countries.

Despite these limitations, our large study provides valuable new scientific insights. Its size, including more than 4 million notified cases, suggested that our findings are robust and valid. Absence of evidence for a significant contribution of cross-border transmission to in-country transmission in our study area lends support to voices questioning the effectiveness of border closures in mitigating domestic transmission. This is particularly important, given the sizeable proportion of the EU population who live in these areas, and the huge impact borders have in terms of economy, employment, social mobility, and the mental wellbeing of the countries’ citizens. Available GIS research on the transmission of pathogens, including COVID-19, has been limited to a national or subnational level, turning a proverbial blind eye to cross-border areas. Our study is the first of its kind to attempt to fill this scientific gap by assessing cross-border transmission.

Insights from our study align well with the EU Serious cross-border threats Regulation. In response to the COVID-19 pandemic, the EU aims to enhance its legal framework for epidemiological surveillance and response to cross-border health threats. This includes broadening reporting requirements, analysing health system indicators, and improving cooperation between Member States and EU agencies such as the European Centre for Disease Prevention and Control (ECDC), the European Medicines Agency (EMA), and also international organisations, in particular the World Health Organization (WHO), while considering the burden on national authorities [10].

To guide future policy and decision-making, it may be of value to recognise that there was in this study no evidence of substantial cross-border transmission within EU Euroregions. Consequently, our study lends support to a cautious approach regarding the implementation of border closures or movement restrictions, given the considerable social and economic impact carried by these measures. Further research is needed to more accurately assess these measures in terms of their proposed mitigating effects and potential side effects, and to determine whether their implementation can be justified as an effective and proportionate public health measure. However, this requires the availability of comparable, high-resolution data across borders, including genomic surveillance, individual travel histories, mobility patterns, and comparative policy analysis. The lack of harmonized data between countries remains a major barrier to understand the true impact of border-related measures. Coordinated cross-border and regional data infrastructures will be essential to enable more robust evaluations in future pandemics.

Authors contributions

BZ and CH designed and organized the study. BK conducted the analyses. BZ drafted the first version of the manuscript. BK, VH and CH contributed to the final version of the manuscript. All authors thoroughly reviewed and verified the final manuscript to ensure the accuracy and integrity of the content of the final version of the manuscript.

Funding

The data utilized in this study were collected through funding provided by Province of Limburg, the Netherlands (acting also on behalf of provinces Gelderland and Overijssel), Staatskanzlei of the state of North Rhine-Westphalia (NRW), and Dutch Ministry of the Interior and Kingdom Relations (Doc-00116170). The subsequent data analysis was supported by a grant from the national Dutch funding body ZonMW (10710032310028). It is important to note that none of these funding entities had any involvement in the study design, data analysis, interpretation of results, or the decision to publish the findings.

Data availability

The data used in this study were already publicly available through the national institutes of the three involved countries: the National Institute for Public Health and the Environment (RIVM) for the Netherlands, the Robert Koch Institute (RKI) for Germany, and Sciensano for Belgium.

- The official Dutch dashboard can be found at: https://www.rivm.nl/corona/actueel/weekcijfers. The official Belgian dashboard can be found at: https://datastudio.google.com/embed/reporting/c14a5cfc-cab7-4812-848c-0369173148ab/page/ZwmOB.

- The official German dashboard can be found at: https://experience.arcgis.com/experience/478220a4c454480e823b17327b2bf1d4.

Any data-sharing requests should be directed to these national bodies, as the authors do not have the authority to grant access to these datasets.

Declarations

Ethical approval

This study utilized fully anonymized, publicly available data from national public health institutes of the Netherlands (RIVM), Germany (RKI), and Belgium (Sciensano). In accordance with national legislation and institutional requirements, written informed consent was not required for the participation in this study. No additional administrative permissions were necessary to access the data, as it was freely available and complied with existing laws and regulations, including those related to privacy protection. Given that the data used was aggregated, anonymous, and publicly accessible, ethical approval from the Maastricht University Ethics Committee was not deemed necessary.

Consent for publication

Not applicable.

Conflict of interests

The authors declare no competing interests.