Introduction

The COVID-19 pandemic spread rapidly around the world from early 2020. In response, many countries implemented control measures related to international travel with these including border closures, partial travel restrictions, exit or entry screening, and quarantine of travellers1. A systematic review of the effectiveness of these interventions1 reported that: “broadly, travel restrictions may limit the spread of disease across national borders. Entry and exit symptom screening measures on their own are not likely to be effective in detecting a meaningful proportion of cases to prevent seeding new cases within the protected region; combined with subsequent quarantine, observation and PCR testing, the effectiveness is likely to improve”.

Indeed, such measures when combined with various public health and social measures (PHSMs) such as mask use and physical distancing (sometimes including “lockdowns”), have effectively eliminated community transmission of the SARS-CoV-2 pandemic virus in a number of jurisdictions 2. These include those with large land borders such as China and Vietnam, but also islands such as Taiwan, Australia and New Zealand2. A number of small Pacific island nations that have used tight border controls have even completely avoided any community transmission of SARS-CoV-2 (e.g., Samoa, Tonga, Cook Islands).

Despite the apparent economic benefits with elimination strategies in terms of GDP impacts during 2020, when compared to countries using suppression strategies2, there is growing interest in how they can safely open up international travel between them to regain the social and economic benefits of quarantine-free travel (e.g., for family reunions, for business travel and for tourism). Indeed, there are arguments for how progressive expansion of “green zones” might open up the possibility of regional elimination of COVID-19 and then ultimately provide a chance of global eradication3.

Two countries that have had partial quarantine-free travel are Australia and New Zealand, and as of March 2021 they were in the process of expanding on this to make such travel two-way and to involve all Australian States and Territories4. Given such developments, we aimed to model the risk of COVID-19 outbreaks associated with international air travel from Australia to New Zealand, along with the likely impact of various control measures that could be used to minimise the risk of such outbreaks. We also aimed to consider the risk for such international travel from Japan and the United States (US) to New Zealand in scenario analyses.

Methods

Model design and parameters for SARS-CoV-2 and COVID-19

We used a stochastic SEIR type model with key compartments for: susceptible [S], exposed [E], infected [I], and recovered/removed [R]. The model is a stochastic version of CovidSIM which was developed specifically for COVID-19 (http://covidsim.eu; version 1.1) and built in Pascal. The stochastic simulation of the compartmental model followed the procedure described by Gillespie 5. If not stated otherwise, one billion simulations were run per combination of intervention strategies. Each simulation started with the random sampling of the initial infection state of 300 passengers who would then (if infected) progress in their natural history and infect others (see Supplementary Information for further detail). In a separate set of simulations, probabilistic parameter sampling within parameter uncertainty ranges (probabilistic sensitivity analysis) was also conducted (see Supplementary Information).

Work has been produced from previous versions of this model6,7,8, and the equations and their stochastic treatment are detailed in two of these outputs9,10. Code for the model is available online is available online at GitHub (https://github.com/nick-wilson-github/Air-travel-Covid-risk-modelling/tree/main).

The parameters were based on available publications and best estimates used in the published modelling work on COVID-19 (as known to us in February 2021).

Prevalence of infection in Australia reflecting an outbreak

To estimate the prevalence of SARS-CoV-2 infection in Australia as a source country we used historical data on reported cases of infection over the 1 April 2020 to 25 February 2021 period (Table 1). This does not reflect the COVID-19-free status that Australia has achieved subsequently in early 2021, but rather the situation equivalent to a substantial outbreak (actually one involving an average of 74 new cases per day). But in scenario analyses we also considered smaller outbreak sizes. For the simulations, travellers to New Zealand were randomly sampled from the Australian population. In most of our scenarios, travellers underwent pre-flight testing before boarding (see Fig. 1 and details below).

Table 1 Input parameters relating to the countries of origin for the travellers arriving in New Zealand (NZ).
Full size table
Figure 1
figure 1

Flow diagram of the modelled movements of air travellers from Australia (when experiencing an outbreak) including the key interventions (simplified and not showing details around travellers seeking medical attention when symptomatic in New Zealand, isolation of identified cases and contact tracing).

Full size image

Selection of control measures

We identified plausible control measures from the published literature1 and also an online review of strategies identified by an IATA Medical Advisory Group11. These controls are shown in Fig. 1 and Table 3.

Air travel to New Zealand

We simulated one three-hour flight per day from Australia to New Zealand, carrying 300 passengers. A wide range of aircraft were used on this route in the pre-COVID-19 pandemic era, with common ones being the Boeing 777-200 which takes 312 passengers and the Airbus A300-300 with 297 passengers. One flight per day is a small proportion of the level of travel in the pre-pandemic time (i.e., 7.1% of the of 1,542,467 traveller arrivals from Australia to New Zealand prior to the pandemic in the year to January 202012). In additional scenario analyses we also considered flights from Japan and the US (with their different population burdens of infection and longer flight times), albeit with the same assumptions for passengers as per flights from Australia to New Zealand.

In-flight transmission risk

We identified a published review on the transmission of SARS-CoV-2 on aircraft13. Using this and subsequently published literature (see Supplementary Information for details), we estimated the number of hours of exposure to infected cases (number of infected people on the flights x flight hours). From this we identified two in-flight infections arising from 933 exposure-hours, giving an estimated risk of transmission per hour of flying in a plane containing an infectious person of 0.00214 (Table 2).

Table 2 Input parameters used for modelling the potential spread of COVID-19 infections with the stochastic version of CovidSIM (v1.1) with New Zealand as a case study.
Full size table

Arrival in New Zealand

Upon arrival in New Zealand, travellers were either placed in supervised quarantine for various periods of days (current practice in New Zealand is for 14 days) and then released to move freely, or, as an alternative to quarantine, we considered various combinations of PCR testing (on days 1, 3 and 12). Until their last PCR test, we assumed that people could move freely around New Zealand, but were required to attend official facilities for testing and to wear a mask while in the presence of other people (Table 2). We further assumed that half of the passengers who develop symptoms during this period would report these symptoms within one day. Also, we assumed that if passengers are tested positive, or if they reported symptoms themselves, contact tracing would identify 75% of their infected contacts in New Zealand who would be isolated after another delay of one day.

Ongoing infection transmission in New Zealand

Secondary cases were assumed to arise from spread from incoming infected travellers in the community in New Zealand. Tertiary cases were those who were infected by secondary cases before the latter were isolated, and were assumed to have the full length of their infectious period ahead of them. Some of them then can trigger an outbreak.

Control measures assumptions

The full details on the control measures we considered are detailed in Table 3.

Table 3 Control measures used and their estimated efficacy in preventing SARS-CoV-2 transmission.
Full size table

Results

Our baseline results used historical data for Australia (equivalent to a border failure resulting in an outbreak generating 74 new cases/day in an otherwise COVID-19-free Australia) and one flight per day to New Zealand with no interventions in place (other than mandated masks on flights). This resulted in the median time to an outbreak in New Zealand being 0.2 years (95% range of simulation results: 3 days to 1.1 years) or after a mean of 110 flights (Table 4). However, the risk progressively declined with the addition of pre-flight testing, testing when in New Zealand, mask use up to the last test, with symptom reporting, and contact tracing. Given all these (with testing on days 1, 3 and 12 in New Zealand), the median time to an outbreak was extended to 1.5 years (20 days to 8.1 years; or a 37% chance per year; or after a mean of 802 flights) (Table 4). Mask wearing by travellers when in the New Zealand community (up to the time of their last test), had a much larger impact than various aspects of the proportion self-reporting symptoms and contact tracing performance. However, 14 days quarantine was a more effective strategy with a median time to an outbreak of 4.9 years (66 days to 26.2 years; mean of 2,594 flights); albeit combined with a pre-flight test. Even better was 21 days of quarantine which was associated with virtually a zero risk of an outbreak ever occurring.

Table 4 Results of the simulations of the baseline risk (no interventions) and for multi-layered packages of interventions to prevent COVID-19 outbreaks in New Zealand (NZ) (assuming a historical level of infection in Australia that was equivalent to a border failure resulting in an outbreak generating 74 new cases/day (as per Table 1) and mandatory mask use on international flights).
Full size table

Table 5 details a range of scenario analyses including if Australia’s infection prevalence was 10 times lower than in the historical base case (i.e., approximating if Australia had a small outbreak from a border failure with around seven new cases per day). For this particular scenario and with no interventions the annual risk of an outbreak was 28.3% and had a median time to an outbreak of 2.1 years. For an even lower prevalence (representing only two new community cases per week in the whole of Australia), the annual risk of an outbreak with no interventions was 1.2% and had a median time to an outbreak of 56 years. But these risks would increase with higher travel volumes as per the scenario of a 20-fold increase in travel from Australia.

Table 5 Scenario analyses covering different source countries, SARS-CoV-2 infection burdens and flight volumes (with the base case for comparison).
Full size table

The high prevalence of infection for the US meant that, even with quarantine, the median time to an outbreak was only 20 days, or after only a mean of 29 flights (Table 5). But the equivalent values for Japan were much longer (1.7 years) and larger (870 flights).

The stochastic simulations using base case parameters presented in this main text were also directly compared with a probabilistic sensitivity analysis (PSA) involving random sampling from the parameter distributions (results in the Supplementary Material: Table S3). This comparison showed almost identical results from the two approaches.

Discussion

Main findings and interpretation

This analysis examined the risk of COVID-19 outbreaks in a COVID-19-free nation (New Zealand), if there was air travel from a low-prevalence country (e.g., if Australia experienced various sizes of outbreaks from border failure and lost its COVID-19 “elimination status”). Using the historical data for Australia (equivalent to an outbreak with 74 new cases per day) and no interventions, we estimated that there would be such an outbreak of COVID-19 in New Zealand after a median time of only two to three months. Fortunately, the multi-layered packages of interventions we modelled reduced this risk to much lower levels. Indeed, without quarantine, the use of a package of measures (pre-flight testing, PCR testing in New Zealand, mask use and contact tracing) could reduce the risk to potentially tolerable levels—if health authorities had confidence in the application of these measures in the real world (e.g., adherence with mask use by travellers and minimal defaulting on testing when in the New Zealand community).

More realistically however, for the situation in early 2021 where Australia has effectively eliminated COVID-19 in the community, is to consider the smaller outbreak scenarios. These would suggest a low risk of an outbreak in New Zealand (e.g., only a 1.2% annual risk with no interventions and a small outbreak of two new cases per week in Australia). But even at this low risk level, New Zealand health authorities might still wish to promote existing digital tools to incoming travellers so as to facilitate rapid outbreak control (e.g., New Zealand encourages QR code scanning when entering buildings and buses etc.).

But for travellers from Japan, the risks are much higher than for travellers from Australia and quarantine would probably remain appropriate. Whereas for travellers from the US the very high risk might suggest that tighter travel restrictions are more appropriate until epidemic spread was reduced. Alternatively processes such as pre-flight quarantine could be considered for US travellers.

Our findings have some compatibility with those of another modelling study which reported that various interventions (including pre-flight and post-flight testing and five day quarantine) would reduce spread of SARS-CoV-2 associated with domestic travel within the US29. But the package of interventions modelled were less intensive than in our model and at best reduced the number of “infectious days” in the modelled cohort by only 70%.

It is likely that all these travel-related risks will decline as vaccinations are provided to: (i) the population of source countries; (ii) travellers (in the weeks prior to travel); and (iii) the population of recipient countries. As such, future modelling should consider vaccination coverage and vaccination effectiveness in preventing transmission. Future modelling should also ideally factor in costs so that cost-effectiveness ratios can be calculated for various intervention packages and their marginal adjustments. The relevance of these costs to policy-makers might also be impacted by who is paying. For example, if all incoming travellers were charged a COVID-19 levy and aspects of the system could be made user-pays (indeed some charges are already used in the New Zealand border system for quarantine).

Ultimately, there is also a need for full cost–benefit analyses which consider the benefits of increased travel to the recipient society and economy—along with the risks of outbreaks that need to be rapidly controlled or else pose a risk of lockdown measures being required.

Study strengths and limitations

This is the first such study (that we know of), to model such interventions in the context of preventing the re-introduction of SARS-CoV-2 into a country that has eliminated it. We were also able to consider a wide range of control interventions and to package these in multiple layers of defence. Nevertheless, there is quite high uncertainty around some of the parameters we used. For example, the prevalence of infection within source countries is, in reality, highly heterogeneous (by age group, social group and locality) and will vary over time. Indeed, Australia in early 2021 had effectively eliminated SARS-CoV-2 transmission with just occasional small and rapidly controlled outbreaks arising from border control failures23. Furthermore, our data on the effectiveness of masks on aircraft was based on a limited amount of real world experience (i.e., only eight flights with cases on board and mask mandates in place). We also did not model infection amongst air crew due to the complexity of their international movements and different control measures used with this occupational group, but note that these personnel have been a rare cause of COVID-19 related border failures in New Zealand to date.

Another limitation is that we did not account for a small proportion of travellers who might cancel their flight to New Zealand if becoming symptomatic after infection with SARS-CoV-2. Furthermore, the probability of symptomatic illness amongst travellers will probably be different than in the general population (e.g., if travel is dominated by younger adults who are less concerned with pandemic risks). We also assumed full adherence to testing regimens within the New Zealand community, though potentially this could be achieved if large fines were imposed or if travellers paid a large financial bond that was only redeemed after full adherence.

Finally, we assumed quarantine only failed due to a tiny proportion of cases having very long incubation periods. In reality, however, a country like New Zealand has had failures with its COVID-19 quarantine system with facility workers and other travellers becoming infected23. This is because it uses re-purposed hotels instead of purpose-built facilities and does not confine the travellers to their rooms (i.e., there are shared corridors, lifts, exercise areas and smoking areas). Given all such issues and ongoing improvements in knowledge of the transmission dynamics of SARS-CoV-2, this type of modelling work should be regularly revised and be performed using different types of models.

Conclusions

This modelling study suggests that the risk of an outbreak in a previously COVID-19-free country is extremely dependent on the source country of the incoming travellers. In the situation of Australia experiencing a large outbreak, the risk could potentially be reduced to tolerable levels with a package of multi-layered interventions (particularly with repeated testing and mask use) and no quarantine. Nevertheless, quarantine is likely to remain important where the source country has high disease burdens. However, all approaches require public and policy deliberation about acceptable risks, and continuous careful management and evaluation.