The impact of air travel on the precocity and severity of COVID-19 deaths in sub-national areas across 45 countries

Abstract

Human travel fed the worldwide spread of COVID-19, but it remains unclear whether the volume of incoming air passengers and the centrality of airports in the global airline network made some regions more vulnerable to earlier and higher mortality. We assess whether the precocity and severity of COVID-19 deaths were contingent on these measures of air travel intensity, adjusting for differences in local non-pharmaceutical interventions and pre-pandemic structural characteristics of 502 sub-national areas on five continents in April-October 2020. Ordinary least squares (OLS) models of precocity (i.e., the timing of the 1st and 10th death outbreaks) reveal that neither airport centrality nor the volume of incoming passengers are impactful once we consider pre-pandemic demographic characteristics of the areas. We assess severity (i.e., the weekly death incidence of COVID-19) through the estimation of a generalized linear mixed model, employing a negative binomial link function. Results suggest that COVID-19 death incidence was insensitive to airport centrality, with no substantial changes over time. Higher air passenger volume tends to coincide with more COVID-19 deaths, but this relation weakened as the pandemic proceeded. Different models prove that either the lack of airports in a region or total travel bans did reduce mortality significantly. We conclude that COVID-19 importation through air travel followed a 'travel as spark' principle, whereby the absence of air travel reduced epidemic risk drastically. However, once some travel occurred, its impact on the severity of the pandemic was only in part associated with the number of incoming passengers, and not at all with the position of airports in the global network of airline connections.

Figure 1

The drop in global air traffic induced by COVID-19. Panel (A) Shows the number of incoming passengers by subnational areas in April 2019; panel (B) shows the situation one year later in April 2020. Panel (C) shows the average change in the monthly number of air passengers in 2020 compared to 2019 (with a value smaller than 1 indicating a drop in passengers). Panel (D) shows the share of region-pairs that saw no drop in air passenger traffic in 2020 compared to the same month in 2019. A drop is defined here as the number of passengers reaching 90 per cent or less of the value in the previous year (assuming that a drop of less than 10 per cent is not meaningful and may simply indicate random fluctuation). Source: Sub-National COVID-19 Incidence and Determinants Dataset.

Figure 2

The impact of air passenger traffic on the severity of COVID-19 deaths (number of deaths per week) in April–October 2020, controlling for population mixing (NPI), structural predispositions and recursive effects. Generalized linear mixed-effects models. Risk ratios and confidence intervals. For full model specification with coefficients, see Table C1 in Supplementary Information. Source: Sub-National COVID-19 Incidence and Determinants Dataset.

Figure 3

The impact of air passenger traffic on the severity of COVID-19 deaths (number of deaths in the first week of April–October 2020), controlling for population mixing (NPI), structural predispositions and recursive effects. Negative binomial regressions with continent fixed effects. Risk ratios and confidence intervals, For full model specifications with coefficients, see Table C2 in Supplementary Information. Source: Sub-National COVID-19 Incidence and Determinants Dataset.

Figure 4

The impact of air passenger traffic on the severity of COVID-19 deaths in the first 30 epidemiological weeks (week1 = first COVID-19 death) controlling for population mixing (NPI), structural predispositions and recursive effects. Negative binomial regressions with fixed effects by continent. Risk ratios and confidence intervals. For full model specifications with coefficients, see Table C4 in Supplementary Information. Source: Sub-National COVID-19 Incidence and Determinants Dataset.