The impact of earlier reopening to travel in the Western Pacific on SARS-CoV-2 transmission

Highlights

• •Earlier relaxation of border constraints in the Asia-Pacific region could have been possible.
• •Case counts would have increased slightly for countries with large COVID-19 outbreaks.
• •Selective border reopening and on-arrival testing greatly reduce importation risk.

Abstract

Background

The COVID-19 pandemic has led to a fall of over 70% in international travel, resulting in substantial economic damages. The impact is especially pronounced in the Asia-Pacific region, where governments have been slow to relax border restrictions.

Methods

A retrospective approach was used to construct notional epidemic trajectories for eight Asia-Pacific countries or regions, from June to November 2021, under hypothetical scenarios of earlier resumption of international travel and selective border reopening. The numbers of local infections and deaths over the prediction window were calculated accordingly.

Results

Had quarantine-free entry been permitted for all travellers from all the regions investigated, and travel volumes recovered to the 2019 levels, Australia, New Zealand, and Singapore would have been the three most severely affected regions, with at least doubled number of deaths, while infections would have increased marginally (< 5%) for Japan, Malaysia, and Thailand.

Conclusions

Earlier resumption of travel in Asia-Pacific, while maintaining a controlled degree of importation risk, could have been implemented through selective border-reopening strategies and on-arrival testing. Once countries had experienced large, localized COVID-19 outbreaks, earlier relaxation of border containment measures would not have resulted in a great increase in morbidity and mortality.

Introduction

Globally, COVID-19 has had a substantial influence on cross-border travel, reducing the volume of international tourist arrivals to 30% of the pre-pandemic level in 2020 and 2021 [1]. This plunge can largely be explained by travel restrictions implemented from early 2020 to reduce disease importation. Such border-control measures have substantially lowered importation risks in settings where they have been carefully implemented [2], most notably in the upper-middle- and high-income countries of Asia-Pacific, where border control was coupled with measures to control local spread until vaccines became available, thereby keeping mortality rates low [3].

The economic and social consequences of travel bans have, nevertheless, been extensive. The drastic reduction in air traffic led to an estimated loss of US$370 billion for airlines in 2020, of which 32% was borne by Asia-Pacific countries, and losses of over US$100 billion for airports and air navigation service providers [4], affecting millions of jobs in aviation around the globe [5]. Tourism was also adversely affected, with the top ten source markets, making up nearly half of all international travelers, advising against non-essential travel abroad [6].

Countries in Asia-Pacific — the initial pandemic epicenter — have implemented different policies for border control, but frequently these have restricted who may enter the country, with many foreigners being prohibited from entry, and travelers being subject to forms of quarantine, as well as pre-departure and post-arrival tests [7], [8], [9]. By the end of November 2021, before the global prevalence of the variant B.1.1.529 (henceforth, the Omicron variant), two thirds of destinations in this region had still completely closed their borders to international tourists [10]. Such blanket travel bans were, however, not supported by WHO, even following identification of more transmissible variants of concern [11].

As vaccination programs were rolled out, policies to resume intraregional travel were introduced. Singapore, for example, launched several Vaccinated Travel Lanes [12] in late 2021 to allow vaccinated visitors from countries or territories with low transmission risk to enter without quarantine if they provided a negative on-arrival test. Similar quarantine-free policies were also adopted by Australia and New Zealand for travelers between these two countries [13]. Nevertheless, the expansion of international travel in the Asia-Pacific region has been much more cautious than in other parts of the world, even as autochthonous transmission of more transmissible variants became widespread.

The question therefore arises as to whether the travel restrictions that were effected in Asia-Pacific early in the course of the pandemic, and the concomitant economic and social costs, were retained for an excessive length of time, and therefore the degree of relaxation that would have maintained low disease prevalence could have been greater. In this study, utilizing the estimated infection potential of SARS-CoV-2 based on prevailing within-country control policies, we quantified the impact of easing border control on local transmission in a collection of countries in the Western Pacific, while exploring hypothetical scenarios and assessing the feasibility of earlier reopening.

Materials and methods

Sources of data

The study focused on eight countries or territories (henceforth, regions) in the Western Pacific: Australia (AU), Japan (JP), the Republic of Korea (KR), Malaysia (MY), New Zealand (NZ), Singapore (SG), Thailand (TH), and Taiwan (TW). These eight were chosen because they are all upper- or upper-middle-income regions that imposed border restrictions, including to each other, despite in many cases having little autochthonous transmission of COVID-19. The study period was chosen as June to November 2021, when the Delta variant dominated but before the Omicron variant emerged [14,15]. Daily counts of COVID-19 cases and deaths were obtained from either the Johns Hopkins University Covid-19 database [16] or authoritative websites of the eight regions (Figure S1). Monthly incoming visitor data and population data were collected from official statistical databases of the eight regions (Figure S2), from which the connectivity from region A to B at time $t$, $c_{t,AB}$, was calculated as the (average) proportion of residents of A that arrived at B on day $t$ (see Supplementary Information). Sources of the case counts, international arrival data, and population data are also listed in the Supplementary Information (Tables S1–S3).

Statistical analysis

Estimation of real-time infection potential

The real-time infection potential was characterized using the time-varying effective reproduction number $R_t$, defined as the ratio of new cases generated by the total infectiousness of infected cases at time $t$ [17], using the EpiFilter algorithm (details in the Supplementary Information) [18]. This was assumed not to change under other scenarios.

Simulating new incidence curves

To model infection sizes under scenarios in which travel restrictions were eased, notional epidemic curves were constructed using different connectivity matrices derived from the travel data, assuming the transmission potential of imported cases was the same as that of local cases.

Let $Y_{t,i}$ and ${\tilde{Y}}_{t,i}$ be the observed and notional case counts for region $i$ at time $t$, respectively. To account for the possible effects of herd immunity, if $R_{t,i}$ was the estimated effective reproduction number for region $i$ at time $t$, the corresponding effective reproduction number ${\tilde{R}}_{t,i}$ for the notional scenario was:

$${\tilde{R}}_{t,i}=\frac{N_i^{2021}-r\times {\sum }_{s=1}^{t-1}{\tilde{Y}}_{s,i}}{N_i^{2021}-r\times {\sum }_{s=1}^{t-1}{Y}_{s,i}}\times R_{t,i},$$

where $N_i^{2021}$ is the population size of region $i$ in 2021, and $r$ is a scaling factor accounting for under-reporting of COVID-19 infections, which was set at 1.6 — the ratio between the increase in seroprevalence and reported percentage infected from May to December 2021 in the USA [16,19]. Additionally, a sensitivity analysis was conducted with different $r$ values involved in the model, but it is found that different $r$s had a limited impact on the final infection size (Figure S5, Table S5).

Following Cook's method of constructing the effects of alternative intervention strategies [20], the same stochasticity was assumed for both actual and notional epidemic trajectories, by fixing the cumulative distribution functions for each datum. If $q_{t,i}$ is the probability that the case count in region $i$ on day $t$ is no larger than $Y_{t,i}$, i.e.

$$q_{t,i}=F_{t,i}\left(Y_{t,i}\right)=P_{t,i}\left(X_{t,i}\le Y_{t,i}\right),$$

where $F_{t,i}(·)$ is the cumulative density function for the random variable $X_{t,i}$ that follows a Poisson distribution with parameter (mean and variance) ${\lambda }_{t,i}=R_{t,i}{\sum }_{d=1}^p{w}_d{Y}_{t-d,i}$ and $w_{1:p}$ is the corresponding serial interval distribution, the number of new cases reported on day $t$ for region $i$ in the notional outbreak is:

$${\tilde{Y}}_{t,i}={\tilde{F}}_{t,i}^{-1}\left(q_{t,i}\right).$$

In this, ${\tilde{F}}_{t,i}(·)$ is the cumulative density function for the Poisson distribution with parameter ${\tilde{\lambda }}_{t,i}={\tilde{R}}_{t,i}{\sum }_{d=1}^p{w}_d({\tilde{Y}}_{t-d,i}+{\sum }_{t_1=t-d}^{t-1}{\tilde{Z}}_{t-d,i,t_1})$, while ${\tilde{R}}_{t,i}$ is the time-varying effective reproduction number at time $t$ for region $i$ in the hypothetical scenario, and ${\tilde{Z}}_{t,i,t_1}={\left({\mathit{C}}_{t_{1,}·i}\right)}^T·{\stackrel{\sim }{\mathbf{Y}}}_t={\sum }_{j=1}^n{c}_{t_1,\mathit{ji}}\times {\tilde{Y}}_{t,j}$, the expected number in the notional outbreak of foreign cases infected at time $t$ and imported to region $i$ at time $t_1\ge t$.

${\stackrel{\sim }{\mathbf{Y}}}_t={({\tilde{Y}}_{t,1},{\tilde{Y}}_{t,2},\cdots ,{\tilde{Y}}_{t,n})}^T$ is defined as the vector of differenced notional case counts for the selected regions at $t$, and ${\mathit{C}}_t=\left(c_{t,\mathit{ij}}\right)$ as the time-varying connectivity matrix with zero principal diagonal elements and the $\left(i,j\right)$th element $\left(i\ne j\right)$ representing the proportion of residents in location $i$ who visited location $j$ ($j\ne i$) at time $t$, quantifying the travel flow from region $i$ to region $j$ at that time. Owing to the discreteness of Poisson distributions, the solution $x=F^{-1}\left(q\right)$ was deemed the smallest integer such that $P\left(X\le x\right)\ge q$.

Assuming an average of 21 days from notification to death [21], the region-specific, time-varying case fatality rate, $CF{R}_{t,i}$, was obtained by smoothing the ratio of reported deaths on day $\left(t+21\right)$ to reported case counts on day $t$ for region $i$, using the mgcv R package [22]. The number of expected additional deaths for region $i$ that would have resulted under the alternative scenarios was calculated as:

$$\sum _{t}CF{R}_{t,i}\times ({\tilde{Y}}_{t,i}-Y_{t,i})$$

which was converted to additional deaths per million population for comparisons between regions.

Scenarios

Travel data from 2021 were used to establish the baseline scenario that visitors travelling among these eight regions were able to move freely in local communities upon their arrival, i.e. not subject to quarantine, and the impact on the number of local cases from June 1 to November 30, 2021 was estimated.

Visitor and population data in 2019 were then used to model how COVID-19 transmission would be if intraregional travel flows recovered to pre-pandemic levels, which were assumed to match the corresponding period of 2019. For this, different travel-resumption dates were considered (the first day of each month from June to November).

Next, the case in which each region selectively opened its border only to some of the other seven regions was considered. If region $i$ were open to region $j$, all visitors from $j$ would be allowed to enter $i$ without any movement restrictions upon arrival, while travel restrictions would remain unchanged for visitors from the regions $i$ was not yet open to (i.e. they would still be subject to quarantine and have negligible effects on local transmission). For each region, this began with the lowest risk region it could open to (see Supplementary Information), under which scenario infections from June to November 2021 were simulated. Then, the set of regions to which its borders were open was successively expanded. It was assumed that increasing infections under these scenarios did not cause secondary effects on other regions, since the effects of increased travel transpired to be relatively small.

Finally, three strategies with varying degrees of resumption of quarantine-free entry for visitors from the eight regions were considered:

• S1. Without an entry test.

• S2. Following a negative rapid antigen test (RAT).

• S3. Following a negative polymerase chain reaction (PCR) test.

The false negative rates for RAT and PCR, averaged over the infectious period, were assumed to be 40% and 10%, respectively [23,24]. Thus, the expected numbers of imported cases that could affect the future local incidence curve were 100%, 40%, and 10% of the total imported cases. The simulation interval for all three strategies was from June to November, 2021.

Results

Estimated $R_t$ values from June to November, 2021 for the eight regions are presented in Figure S3.

Impact on local cases of cancelling movement restrictions for foreign visitors

Without travel restrictions, the epidemic trajectory for New Zealand would have been fundamentally altered by imported cases, with an epidemic wave of at least 1000 daily infections expected in November 2021 and a total case count over the 6-month period of 11.5 (95% credible interval [CrI] 8.0–16.7) times the observed value. Australia and Singapore would also have experienced larger outbreaks from August and September to November 2021, with over 500 additional local infections per day, and 1.4 (95% CrI 1.3–1.5) times the overall number of cases by November. The other regions, by comparison, would not have been so adversely affected by importation, which would have brought fewer than 6000 additional cases in total for each region (Figure 1, Figure S6).

Figure 1.

Local cases averted by border containment measures in 2021. Comparisons between daily case counts with (‘observed’ — dots in blue for reported case counts) and without (‘notional’ — in red, with line for mean and shade for 95% CrI) implementation of travel restrictions (i.e. quarantine requirement for foreign visitors after their arrival) from June to November 2021, assuming the numbers of foreign visitors were the same as those in the real situation. For presentation purposes, observed incidence is presented for every third day.

Cases increased due to ease of travel restrictions and recovery of tourism

Diverse tourism recovery dates

Changes in infection incidence that would have been caused by increases in international arrivals varied greatly from region to region for any fixed tourism recovery time. Had all foreign visitors not been quarantined upon arrival, Taiwan and Singapore would have been the most adversely affected by resumption of tourism to the 2019 level, where, respectively, the total numbers of infections over the 6-month period would have been 2.18 (95% CrI 1.8–2.6) and 1.34 (95% CrI 1.25–1.44) times higher than in the scenario in which the volume of tourists remained at 2021 levels. By comparison, the influence of the rise in incoming tourists would not have been so significant in New Zealand, because lifting border restrictions would have caused surges in infections even when there were not so many international visitors (Figure 2, Figure S7). For Japan, Malaysia, and Thailand, an earlier resumption of tourism in June 2022 would have caused a less than 5% increase in total cases. The relationships between increases in infection sizes and recovery times, however, were non-linear for all eight regions, while a later recovery of tourism did not necessarily mean fewer additional infections per month (Figure S8).

Figure 2.

Infection size comparisons for scenarios with diverse tourism recovery dates. Comparisons (mean and 95% CrI) of infection sizes (from the start of tourism resumption to the end of November) between scenarios with different tourism resumption dates, ranging from the first day of June to November. The reference scenario assumes numbers of foreign visitors remained the same as those in 2021 throughout the 6-month simulation window. In each scenario, no movement restriction was imposed on foreign visitors upon their arrival.

Selective border reopening

Selective reopening to low-incidence regions was then considered, leading to some but not all intraregional travel flows recovering to pre-pandemic levels. Over the period considered, as the number of regions reopened to rose from one to seven, the number of infections would have surged. Outbreak sizes in New Zealand were the most sensitive to border reopening, where visitors from Taiwan, the lowest-risk region it could open to, meant the total infections over the half year would have expanded by 13 (95% CrI 9\05518) times. Other regions, by comparison, would not have been so greatly influenced by importation of cases, and withdrawing border restrictions to New Zealand and Taiwan would have increased case counts by less than 50% elsewhere. Furthermore, for Japan, Malaysia, and Thailand, reopening to all the regions would have increased total infection sizes by no more than 5%. The impact of additionally reopening to regions with large infection and mortality rates, such as Thailand and Malaysia, nevertheless, would have been relatively significant. In particular, reported cases would have increased by at least a quarter over the 6-month period if residents in Malaysia were eligible for quarantine-free entry elsewhere (Figure 3, Figure S9).

Figure 3.

Infection size comparison for borders reopening to different regions. Times of total infections (mean and 95% CrI) for each region by the end of the 6-month simulation period (from June to November 2021), compared with the real situation, under selective opening of borders to some of the other regions (from one to seven, ordered by risk), assuming communication among the regions had recovered to pre-pandemic (2019) levels throughout this period.

Impact on mortality of easing travel restrictions

Singapore was estimated to experience the biggest rise in mortality due to the relaxation of quarantine requirements, with an average increase of 225 deaths per million residents if visitors from the other seven regions were allowed to enter without quarantine (Table 1). This would be followed by Australia and New Zealand, where 60 and 54 additional deaths per million residents would be expected, respectively. The introduction of mandatory testing for international arrivals, however, would have approximately halved the increase in deaths in Singapore and Australia; the reduction was not so great for New Zealand. For the other regions, none of the three proposed strategies would have led to more than 35 more deaths per million residents, and testing upon arrival brought down the fatalities considerably, with fewer than half the deaths of a less cautious exit strategy (Table 1, Table S6, Figure S10).

Table 1.

Deaths per million residents for different travel-resumption strategies. Actual and expected increases in total deaths per million residents for each region over the 6-month period (June–November, 2021), when quarantine-free entry was allowed for all (S1), RAT-negative (S2), and PCR-negative (S3) travelers from the eight Western Pacific regions, assuming communication among the regions had recovered to pre-pandemic levels (the corresponding period in 2019).

Region	Actual	S1: quarantine free		S2: RAT test		S3: PCR test
		Increase	Total	Increase	Total	Increase	Total
Australia	42	60 (140%)	102	33 (78%)	75	21 (50%)	63
Japan	42	1.5 (3.6%)	44	0.5 (1.2%)	43	0.1 (0.2%)	42
Republic of Korea	33	4.5 (14%)	37	1.8 (5.5%)	35	0.7 (2.1%)	33
Malaysia	845	32 (3.8%)	877	9.2 (1.1%)	854	1.6 (0.2%)	847
New Zealand	4	54 (1500%)	58	42 (1200%)	46	37 (1100%)	41
Singapore	125	225 (180%)	349	148 (120%)	273	78 (62%)	202
Thailand	298	14 (4.5%)	312	5.2 (1.7%)	303	1.3 (0.4%)	300
Taiwan	31	20 (64%)	51	8.7 (28%)	39	3.7 (12%)	35

Discussion

Since early 2020, COVID-19 has caused the most severe social and economic setbacks of the past few decades [25]. While border control measures have proven to be effective in containing the spread [26], the resulting reduction in international travel has gone against the interdependency of modern economies. Forward-looking reopening policies are therefore required, to mitigate the harmful effects of the current pandemic, to get prepared for future epidemic waves, and to avoid exhausting the goodwill of policymakers and the public.

The number of cases arising from importation has been found to be closely associated with local incidence and local virus transmissibility [27]. Our analysis suggests that, in the absence of rigorous travel restrictions, new epidemic waves seeded by imported cases would have been substantially more likely to have taken place in regions with few local cases, compared with those that already had large, localized outbreaks. This vulnerability could partially be explained by the use of an estimated $R_t$ corresponding to transmissibility as events transpired, but which might be reasonably expected to have been lower had infections been markedly greater, given these regions’ otherwise successful policies.

The diversified correlations between monthly increases in infection sizes and tourism resumption times proposed in this study suggested the possibly of earlier reopening in some regions. The similarity in average additional infections per month in some cases supports the notion that travel restrictions that reducing international arrivals only postpone an outbreak rather than preventing it, particularly when a region is in the midst of an epidemic [27]. Meanwhile, border reopening to regions with higher risks would generate a disproportionally large number of additional infections. This would align well with a selective reopening strategy, which is in accordance with WHO's recommendation of risk-based lifting of travel restrictions [28].

In addition, our estimation of infection sizes in the hypothetical scenarios with varying border requirements demonstrated the significance of mandatory on-arrival tests. Though the more accurate PCR tests could better bring down importation impact by half for most of the regions investigated compared with RAT, the former's high false-positive rates after recovery [29] would cause unnecessary quarantine for recovered cases who are no longer infectious, and thus extra financial burdens for individual travellers as well as the authorities. In comparison, RAT tests, despite their relatively low sensitivity, are more affordable and give more timely results [30].

Limitations of this study include the utilization of uniform, constant reporting rates under the assumption of sufficient access to COVID-19 testing in the regions considered, but the rates could be region-specific or time-varying, affected by policies adopted, such as the home-recovery programme in Singapore introduced in October 2021 [31]. $R_t$ was estimated under the optimal assumption of minimal leaking risks in quarantine, realized through stringent border restrictions and potent test-trace-isolate strategies [32], while changes would possibly arise with the easing of border restrictions, since the transmission potential might be subjected to exogenous factors, such as mobility patterns and the containment measures implemented [33]. $R_t$ is also likely to be affected by imported cases, particularly when the composition of circulating variants differs from that of local infections and disparities exist in the transmissibility of different variants.

Another simplified assumption was the uniform probability of getting infected for all individuals in the same place, be they foreign visitors, local residents, or people planning to travel abroad. The possibility of reinfection was not considered; nor was the heterogeneity in individuals’ transmission capacity, which might be influenced by symptoms displayed during the infectious period [34,35]. Meanwhile, simulated infection sizes could be affected by the uncertainty in $R_t$ posterior estimates, resulting from the scarcity of case information in regions like New Zealand and Taiwan. This impact could not be eliminated, although it was greatly mitigated by EpiFilter, which provides more precise $R_t$ estimations in low-incidence scenarios compared with other prevailing methods. Furthermore, due to limited data accessibility, only the impact of foreign arrivals was modelled, while that of the residents returning from abroad was not explicitly accounted for, which was likely to double the expected additional infections if taken into consideration.

Finally, since the estimation window for this study was mostly a period when the Delta variant was the dominant strain, parameters employed for prediction were relatively consistent over the period, but the emergence of new variants may modify transmission patterns of the virus, as well as efficacy of vaccination and fatality rates [36,37]. Therefore, the risks of lifting border restrictions assessed in this study might not reflect the situation after the emergence of the Omicron variant, which is more transmissible yet. Nevertheless, the results suggest that the resumption of international travel could have been effected sooner within the Asia-Pacific region, once widespread local transmission had begun.

Conflicts of interest

All authors declare no conflicts of interest.