Geographic disparities in COVID-19 infections and deaths: The role of transportation.

The US government imposed two travel restriction policies to prevent the spread of the COVID-19 but may have funneled asymptomatic air travelers to selected major airports and transportation hubs. Using the most recent JHU COVID-19 database, American Community Survey, Airport and Amtrak data form Bureau of Transportation Statistics from 3132 US counties we ran negative binomial regressions and Cox regression models to explore the associations between COVID-19 cases and death rates and proximity to airports, train stations, and public transportation. Counties within 25 miles of an airport had 1.392 times the rate of COVID-19 cases and 1.545 times the rate of COVID-19 deaths in comparison to counties that are more than 50 miles from an airport. More effective policies to detect and isolate infected travelers are needed. Policymakers and officials in transportation and public health should collaborate to promulgate policies and procedures to protect travelers and transportation workers from COVID-19.

Introduction

The impact of the COVID-19 pandemic has varied substantially by geography (NYT, 2020). The nation first saw cases and deaths on the West Coast, specifically in San Francisco and the Seattle metropolitan areas, WA (Holshue et al., 2020). However, the pandemic quickly exploded in the Northeast where the New York metropolitan area became the epicenter of the disease. This persisted for a few months as Governors from Virginia to Massachusetts struggled to guide their states to flatten the infection and hospitalization curves and minimize the number of deaths. At the same time, other areas of the country seemed to be unaffected by COVID-19 with relatively few cases and deaths despite being major population centers. With the exception of pockets of outbreaks, large areas of some states were seemingly untouched by COVID-19. Even within states with large outbreaks, the number of cases and deaths were not evenly distributed across the state. For example, downstate New York was hit much harder that upstate New York (Thomas, 2020). Other major cities, like New Orleans, Chicago and Detroit stood out like outliers in their respective regions.

To prevent the introduction and spread of COVID-19 in the United States, the US Government issued a series of travel arrival restrictions (DHS, 2020a). Travelers from selected countries were subject to enhanced arrival protocols. Specifically, travelers from these countries were screened for symptoms of COVID-19. These travelers were restricted to landing at selected airports where enhanced public health measures were in place to identify travelers who exhibited overt signs of illness. In addition, carriers were instructed to report ill travelers to appropriate public health officials for evaluation. On January 17, 2020, travelers from Wuhan, China were subjected to enhanced health screening at three US airports: San Francisco (SFO), New York (JFK) and Los Angeles (LAX). On February 2, 2020, this travel restriction was expanded to include all flights carrying persons who have recently traveled from or were otherwise present in the People's Republic of China (CDC, 2020a). These travelers were only allowed to land at one of the eleven airports designated by the Department of Homeland Security where enhanced public health services and protocols were being implemented (DHS, 2020b). A month later on March 4, travelers from the Islamic Republic of Iran became subject to these arrival restrictions. On March 17, travelers from the 26 countries of the European Schengen area (DHS, 2020c) and on March 19, travelers from the United Kingdom or the Republic of Ireland were also restricted to landing in selected airports. The airports where travelers were directed to were:

John F. Kennedy International Airport (JFK), New York;

Chicago O'Hare International Airport (ORD), Illinois;

San Francisco International Airport (SFO), California;

Seattle-Tacoma International Airport (SEA), Washington;

Daniel K. Inouye International Airport (HNL), Hawaii;

Los Angeles International Airport (LAX), California;

Hartsfield-Jackson Atlanta International Airport (ATL), Georgia;

Washington-Dulles International Airport (IAD), Virginia;

Newark Liberty International Airport (EWR), New Jersey;

Dallas/Fort Worth International Airport (DFW), Texas;

Detroit Metropolitan Airport (DTW), Michigan.

In March 17, 2020; Boston Logan International Airport (BOS), Massachusetts; and Miami International Airport (MIA), Florida were added.

In addition to restricting where travelers could enter the country, the US Government issued several Presidential proclamations restricting who could enter the country (CDC, 2020c). However, these proclamations had a number of exceptions including US citizens, permanent residents, and persons related to US citizens. These persons could only enter the country at one of selected 15 airports (Lake, 2020). While these airports were supposed to screen travelers, many travelers entered the country without being screened (Kanno-Youngs, 2020).

These two travel restriction policies meant to prevent the introduction and spread of COVID-19 are probably responsible for the geographic disparities in COVID-19 cases and deaths. The policies may have effectively identified and isolated travelers with symptoms, but they also funneled asymptomatic carriers of the COVID-19 to selected airports; and then allowed these asymptomatic travelers to travel to other airports via connecting flights. This funneling of symptomatic and asymptomatic travelers exposed persons who work at these airports to COVID-19. Unfortunately, airport workers were not provided personal protective equipment and the airports were not routine sanitize for possible contamination. This lack of precaution may have increased the risk of community spread of COVID-19 among residents who lived near these airports.

Consequently, early in the pandemic, persons who were in close proximity to major airports would be at greater risk of infection and mortality. To explore this hypothesis, this study estimates the associations between the proximity to airports and train stations, and the reliance on public transportation to the number of COVID-19 cases and deaths. We hypothesize that because the virus spreads to communities in part through air travelers and related transportation systems, communities closest to transportation hubs will be affected first. Also, these communities will have more cases and deaths during the first phase of the pandemic.

Data and methods

Data and sample size

This is a county level analysis that includes 3,132 counties in the US. In the United States, counties are the primary local legal division in most states. Most counties have governmental units, whose powers and functions vary from state to state (Census, 2020). County governments are typically responsible for locally-based public services including education, police, fire, zoning and public health. There are 3,141 counties and county equivalents in the 50 States and the District of Columbia. On average a county has 104,000 population which varies by urban, suburban and rural areas. In some urban counties in populous states like California, New York and Texas have millions of residents. However, in these same states, there are rural counties with fewer than 25,000 residents. About 89% of counties have some kind of airport, but if we consider major airports only 25% of counties are near a large or mid-size airport.

This study used publicly available data from the Johns Hopkins University (JHU) database (JHU, 2020). This data set provides the daily numbers of confirmed COVID-19 cases and deaths at the county level since January 21, 2020, which is the first day for which COVID-19 data is available from the JHU database. To understand the association between transportation and COVID-19 exposure we merged JHU's data with the Bureau of Transportation Statistics' Airports database (BTS, 2020), Amtrak Station database (BTS, 2020), Census-2018 commuter data, and the American Community Survey-2018 (Census, 2018). The Airports database is a geographic point database of official operational aerodromes in the United States and U.S. Territories that includes the county in which each aerodrome is located (ACS, 2020).

Variable measurement

The study has two main dependent variables: The number of confirmed COVID-19 1) cases and 2) deaths at the county level between January 21, 2020 and August 18, 2020 in the US.

Independent variables

The first set of main independent variables includes distance to the nearest airport, airport volume, number of airports, number of train stations, and use of public transportation at the county level. We also included a dummy variable to identify counties which either contain or are within 25 miles from any of the 13 major airports where incoming international travel was directed. We used the latitudinal and longitudinal geographic points of airports and counties and computed the distance between the centroid of each county and the nearest airport in miles. We then created a categorical variable to classify each county as being located within 25 miles, between 25 and 50 miles, and greater than 50 miles away. We used the number of passengers for each airport in 2019 to categorize them into no volume (no airport), middle volume (1–1.07 million passengers), and high volume (1.07 million −39.9 million passengers) airports. To measure the availability of air and rail transportation, we counted the number of airports and Amtrak intercity train stations in each county (BTS, 2020). Finally, to measure the accessibility of other forms of transportation, we included county-level data on the commute traveling time and the percentage of people who drive alone, carpooled, used public transportation, walked, and any other forms of transportation.

We also used data on the racial composition in terms of percentages of White Non-Hispanic (WNH), Black NH, Hawaiian NH, Asian NH, Native American NH and Hispanic in each county to control for racial disparities in outcomes as reported in many recent publications (Aldridge et al., 2020; Tai et al., 2020; Rodriguez-Diaz et al., 2020).

To control for population and population density, we included the county's population, the number of people per square kilometer, and the average number of people in each household. We also control for the age distribution of population at the county level. Finally, we controlled for the percentage of poverty at the county level.

Statistical analysis

We used descriptive, bivariate, and multivariate analyses to address the associations between COVID-19 cases and death rates and proximity to and size of airports, train stations, and public transportation.

For the main analysis we ran two sets of negative binomial regressions (NBREG) (STATA, 2018) to determine the associations between COVID-19 cases and death rates and proximity to and size of airports, train stations, and public transportation. Negative binomial regression models use the number of occurrences (counts) of an event when the event has extra-Poisson variation, to measure number of cases and deaths that accumulated over time (STATA, 2018). In the negative binomial regression, the measure of effect is reported as incidence rate ratios which is the ratio of the number of incident events of the expressed category to the number of incident events in the base category and represents the impact of an independent variable in terms of a percentage change.

In our main models we controlled for racial/ethnic composition, population, population density, household density, and poverty status at the county level. Errors were clustered at the state level.

We also ran two sets of Cox regressions to explore the time dimension of the relationship between COVID-19 outcomes and transportation. The Cox regression is a series of comparisons of those subjects who fail to those subjects at risk of failing(STATA, 2020). In the Cox regression, the measure of effect is the hazard ratio which represents the risk of failure (i.e., the risk or probability of the event of interest occurring) for observations that survive to a certain time point. The Cox regression model requires that covariates satisfy the proportional hazards assumption, i.e., that the effect of a covariate and the ratio of hazards between two participants is constant across time. Graphically, Fig. 1, Fig. 2 imply that our data satisfies this assumption because the survival curves have similar shapes and do not cross across time (Basu et al., 2004; Hess, 1995). In this analysis we used the Cox regression to estimate the effect of a county's proximity to an airport and the likelihood of getting its first case. Survival in this context implies that a county has yet to receive its first case or death (i.e., the events of interest).

Fig. 1

Survival Curves of First COVID-19 Case and First COVID-19 Death relative to Distance to Major Airports.

Fig. 2

Survival Curves of First COVID-19 Case and First COVID-19 Death relative to Volume of Airports.

In these regressions, we define failure time to be the number of days since January 21, 2020 to either the first case or first death depending on the dependent variable in the model. We similarly control for racial/ethnic composition, population and household density, and poverty status and cluster errors at the state level. Counties without any cases were censored. We used Stata (version 15) for data management and analysis.

Sensitivity analysis

We conducted three sensitivity analyses. First, to ensure that our findings are not just the results of the outbreaks in New York and California, we re-ran the analyses excluding these two major states. The findings were very similar to the original models (See appendices A and B.).

In the second sensitivity analysis we used distance and volume as continuous variables. We found that counties further from an airport or with a lower volume of air travelers had fewer cases and deaths. (See appendix C.)

For the final sensitivity analysis, we ran the original model and censored the data at 113 days (i.e., the midpoint) to learn how associations between COVID-19 cases and death rates and proximity to airports, trains stations, and public transportation changed between the first and the second waves of the outbreak. (See appendices D.)

Results

Cases and deaths

COVID-19 cases and deaths were higher for counties which are closer to an airport. In addition, the numbers of COVID-19 cases and deaths increased as the volume of passengers increased (See Table 1 ). The numbers of deaths and cases were positively correlated with the number of airports, number of train stations, the percentage of adults using public transportation, and the length of commuting time (See Table 2 ).

Table 1

Distribution of COVID-19 cases and deaths by proximity to the airport and volume of passengers travel to the airport.

	Number of COVID-19 (in county)
	Case		Death
	N	Mean (SD)	N	Mean (SD)
Distance to airport
<25 Miles	1124	3684 (13,129)	1124	124 (780)
25-50 Miles	1281	799 (4381)	1281	19 (107)
>50 Miles	727	469 (267)	727	11 (47)
Volume of passengers
No volume	2548	681 (1940)	2548	20 (94)
Medium volume	529	3310 (5477)	529	103 (242)
High volume	55	36711 (49245)	55	1221 (3266)

Sources: JHU-COVID-19 Data (2020 August 21), Airport Data (2019)

Table 2

Associations of COVID-19 cases and death rates and proximity to airports, trains stations and public transportation.

	Correlation
	Number of COVID-19Cases in County	Number of COVID-19Deaths in County
Number of Airport and Train Stations
Num. of Airport at County	0.320***	0.128***
Num. of Train Station at County	0.450***	0.208***
Commute
Drove alone	0.018	−0.027
Carpooled	0.016	−0.024
Public trans	0.368***	0.466***
Walked	0.044***	0.109***
Other	0.088***	0.078***
work from home	0.060**	0.039
Commute Traveling time	0.123***	0.089***
County population	0.837***	0.196***
Population Density	0.151***	0.452***

Sources: JHU-COVID-19 Data (2020 August 21), Airport and Amtrak Station Data (2019) and ACS Data (2018).

We estimated negative binomial regression models to assess the spatial impact of proximity, size and use of the airports, trains, and buses on the extent of the COVID-19 outbreak and report incidence rate ratios. The number of COVID-19 cases and deaths increased with proximity to an airport. Counties closest to an airport (i.e., within 25 miles) had 1.392 (CI: 1.185–1.636) times the rate of COVID-19 cases compared to counties that were more than 50 miles from an airport during the first 215 days of the pandemic. Counties within 25 miles of an airport had 1.545 (CI: 1.234–1.934) times the rate of COVID-19 death compared to counties that were more than 50 miles from an airport during the study period. The volume of passengers coming through an airport was associated with the numbers of COVID-19 cases and deaths. Counties with medium volume airports had 1.284 (CI: 1.058–1.557) times the rate of COVID-19 cases and counties with high volume airports had 0.449 (CI: 0.249–0.809) times the rate of COVID-19 cases and 0.366 (CI:0.177–0.754) times the rate of COVID-19 deaths compared to the counties that had no airport.

Using public transportation and driving alone to work were associated with higher rates of COVID-19 cases and deaths. A one percentage point increase in the county percentage of adults using public transportation is associated with an increase of 1.057 (CI: 1.019–1.095) times the rate of COVID-19 cases and 1.096 (CI:1.048–1.147) times the rate of COVID-19 deaths relative to working from home. The percentage of county residents driving to work was associated higher rates of COVID cases (IRR = 1.030 CI:1.024–1.036) and deaths (IRR = 1.033 CI: 1.025–1.041), respectively (See Table 3 ).

Table 3

Associations of COVID-19 cases and death rates and proximity to airports, trains stations and public transportation.

	Negative Binomial Regression				Cox Regression
	Number of COVID-19Cases in County		Number of COVID-19Deaths in County		Time to First COVID-19Case in County		Time to First COVID-19Death in County
	IRR	CI-95	IRR	CI-95	HR	CI-95	HR	CI-95
Distance to airport: Ref: >50 miles
25> Miles to airport	1.392***	[1.185]-[1.636]	1.545***	[1.234]-[1.934]	1.701***	[1.491]-[1.941]	1.337***	[1.167]-[1.531]
25-50 Miles to airport	1.199***	[1.077]-[1.336]	1.192	[0.980]-[1.451]	1.302***	[1.176]-[1.441]	1.160***	[1.069]-[1.260]
Num. of airport	1.011	[0.994]-[1.027]	1.008	[0.976]-[1.041]	1.001	[0.980]-[1.023]	1.028***	[1.012]-[1.045]
Volume of AirportRef: No volume
Medium volume	1.284*	[1.058]-[1.557]	1.309	[0.980]-[1.748]	1.445***	[1.202]-[1.737]	1.159	[0.926]-[1.450]
High volume	0.449**	[0.249]-[0.809]	0.366**	[0.177]-[0.754]	0.633	[0.340]-[1.182]	1.331	[0.781]-[2.270]
Num. of train stations	0.973	[0.839]-[1.130]	1.049	[0.800]-[1.374]	1.086	[0.989]-[1.192]	1.000	[0.947]-[1.056]
Being in counties or nearby counties with major 13 Airports	0.547*	[0.337]-[0.887]	0.567	[0.273]-[1.179]	1.157	[0.552]-[2.424]	0.941	[0.538]-[1.645]
Being an international airport	1.199	[0.969]-[1.484]	1.214	[0.939]-[1.568]	1.677***	[1.358]-[2.071]	1.464**	[1.163]-[1.843]
CommuteRef: Work from home
Drove alone	1.030***	[1.024]-[1.036]	1.033***	[1.025]-[1.041]	1.011***	[1.006]-[1.017]	1.008***	[1.004]-[1.012]
Carpooled	1.005	[0.986]-[1.025]	0.982	[0.952]-[1.012]	0.984*	[0.970]-[0.999]	0.987	[0.971]-[1.003]
Public trans	1.057**	[1.019]-[1.095]	1.096***	[1.048]-[1.147]	1.089***	[1.056]-[1.123]	1.108***	[1.085]-[1.131]
Walked	0.900***	[0.867]-[0.934]	0.899***	[0.860]-[0.939]	0.964**	[0.940]-[0.989]	0.935***	[0.900]-[0.972]
Other	1.107***	[1.046]-[1.172]	1.049	[0.971]-[1.133]	1.023	[0.978]-[1.070]	1.017	[0.971]-[1.064]
Commute Traveling time	0.999	[0.980]-[1.019]	1.017	[0.991]-[1.045]	1.024**	[1.009]-[1.039]	1.013	[0.999]-[1.027]
County Population
Total county population (/1000)	1.004***	[1.002]-[1.005]	1.004***	[1.002]-[1.005]	1.001***	[1.001]-[1.002]	1.001***	[1.000]-[1.001]
Population density (/100)	0.999	[0.997]-[1.001]	0.998	[0.997]-[1.000]	1.001***	[1.001]-[1.002]	1.001***	[1.001]-[1.002]
N	3132		3132		3084		3084

*p < 0.05, **p < 0.01, ***p < 0.001.

Sources: JHU-COVID-19 Data (Updated August 21, 2020), Airport and Amtrak Station Data (2019), ACS Data (2018)

Notes:

1) All models have been controlled by race/ethnicity groups, population density, number of people at household, age categories and percent of poverty.

2) We have clustered errors by state to capture states policies and interventions.

We estimated a Cox regression to determine how the transportation system was associated with how quickly the COVID-19 virus moved. We calculated the time until the first case and the time until the first death in the county (See Fig. 1). Counties within 25 miles of an airport had a 70.1 percent greater risk of having a COVID-19 case and 33.7 percent greater risk of having a COVID-19 death when compared to counties over 50 miles from an airport. The risk of having COVID cases was greater for counties with a high volume airport compared to counties with no airport. The number of airports increased the hazard of having a COVID-19 death. The proportion of residents using public transportation increased the risk of cases and deaths while carpooling and walking to work decreased the risk of cases.

To better illustrate the impact of the proximity to an airport and the volume of passengers using the airport we graphed the survival curves for these variables. At 56 days, 67.4% of counties within 25 miles of the airport did not have a COVID-19 case, compared to 88.6% of counties between 25 and 50 miles of an airport, and 91.9% of counties greater than 50 miles of an airport. At 126 days, these percentages fell to 1.1%, 5.7% and 14.2%, respectively (See Fig. 1, panel A).

At 70 days, 69.8% of counties within 25 miles of an airport did not have a COVID-19 death, compared to 89.1% of counties between 25 and 50 miles of an airport, and 91.7% of counties greater than 50 miles of an airport. At 126 days, these percentages declined to 27.5%, 49.6%, and 65.1%, respectively. (See Fig. 1, panel A). We see a similar pattern for COVID-19 deaths (See Fig. 1, panel B).

The pattern for passenger volume is similar. At 56 days, 12.0% of counties with high volume airports did not have a COVID-19 case, compared to 40.3% of counties with a medium volume airport, and 94.5% of counties with no airport. At 126 days, these percentages declined to 0%, 1.1%, and 7.1%, respectively. (See Fig. 2, panel A). This pattern holds for COVID-19 deaths. At 70 days, 11.8% of counties with high airports did not have a COVID-19 death, compared to 64.5% of counties with a medium volume airport and 89.1% of counties with no airport. At 126 days, these percentages declined to 3.6%, 24.0%, and 50.5%, respectively (See Fig. 2, panel B). The numbers and rates of COVID-19 cases and deaths were negatively associated with distance from the airport. The fitted lines in appendices E and F illustrate these associations.

Impact of other factors

The estimated associations of transportation system factors and COVID-19 outcomes were robust and independent of other community level factors. Our analysis shows that COVID-19 outcomes were related to other community level factors such as racial composition, household size and population density. We found that a percentage point increase in residents who are Non-Hispanic Native Americans is associated with 1.021 (CI:1.005–1.037) times higher the rate of COVID-19 cases. Similarly, a percentage point increase in residents who are Black Non-Hispanic and Hispanic is associated with 1.025 (CI:1.016–1.033) and 1.017 (CI:1.006–1.028) times higher the number of COVID-19 cases, respectively. Household size is a major predictor. A one-person increase in household size is associated with an increase of 2.538 (CI: 1.948–3.307) times the rate of COVID-19 cases and 2.627 (CI:1.791–3.854) times the rate of COVID-19 deaths relative to working from home. Population density in the community is not found to be associated with the number of cases but is positively associated with the time to the first COVID-19 case and death. (See Table 4 ).

Table 4

Associations of COVID-19 cases and death rates and population density and racial compositions.

	Negative Binomial Regression				Cox Regression
	Number of COVID-19Cases in County		Number of COVID-19Deaths in County		Time to First COVID-19Case in County		Time to First COVID-19Death in County
	IRR	CI-95	IRR	CI-95	HR	CI-95	HR	CI-95
Household Density
Num. of person at HH	2.538***	[1.948]-[3.307]	2.627***	[1.791]-[3.854]	1.249*	[1.023]-[1.525]	1.551***	[1.288]-[1.868]
Racial Composition; Ref. White NH
Hawaiian NH	0.944	[0.755]-[1.181]	0.768***	[0.657]-[0.898]	1.056	[0.939]-[1.188]	0.824**	[0.725]-[0.936]
Asian NH	1.036	[0.985]-[1.089]	1.051	[0.982]-[1.126]	1.035	[0.992]-[1.079]	1.033	[0.999]-[1.068]
Native American NH	1.021**	[1.005]-[1.037]	1.034***	[1.015]-[1.055]	0.994	[0.988]-[1.001]	1.000	[0.992]-[1.008]
Black NH	1.025***	[1.016]-[1.033]	1.034***	[1.024]-[1.044]	1.011***	[1.006]-[1.017]	1.023***	[1.018]-[1.029]
Hispanic	1.017**	[1.006]-[1.028]	1.022***	[1.012]-[1.032]	0.995	[0.991]-[1.000]	1.005*	[1.001]-[1.008]
N	3132		3132		3084		3084

*p < 0.05, **p < 0.01, ***p < 0.001.

Sources: JHU-COVID-19 Data (Updated August 21, 2020), Airport and Amtrak Station Data (2019), ACS Data (2018)

Notes: 1) All models have been controlled by race/ethnicity groups, population density, number of people at household, age categories and percent of poverty.

2) We have clustered errors by state to capture states policies and interventions.

Limitations

This study has a few limitations. One, it is subject to the ecological fallacy. There could be some confounding factors correlated with airports and transportation systems and the number of COVID-19 cases and deaths. We attempt to control for known demographic and economic factors in our models that are associated with the disease. Two, the numbers of COVID-19 cases and deaths are most likely undercounted. Testing in most counties was limited to those persons who had symptoms and presented in a healthcare facility. There may be many undetected cases in counties which would affect our count of cases and measure of time to the first case. However, this is probably not correlated with proximity to or size of the airport. Future research should look at patient-level clinical data to determine the association between air travel and COVID-19 cases and deaths.

Discussion

Proximity to an airport was an important risk factor to COVID-19 infections and death. While efforts to restrict travel were meant to prevent the introduction and spread of the virus, this study suggests that it may have amplified the risks for persons who live near airports. Previous efforts to identify travelers with symptoms and the inability to identify asymptomatic travelers may have heightened the risk for residents of cities near major airports. Our results imply that public health resources for screening, detection, and containment should be initially concentrated in communities around airports of international entry in the early stages of a pandemic. Furthermore, it may also be worthwhile to consider redirecting incoming international air travel to a small number of airports in the country that are farther from population centers and in lower population density areas. Examples would include the use of London Gatwick (LGT) instead of London Heathrow (LHR) in the United Kingdom or the use of Tokyo-Narita (NRT) instead of Tokyo-Haneda (HND) airports in Japan.

Our findings are consistent with recent research by Zhang and colleagues. They showed that frequencies of air flights and high-speed train services out of Wuhan, China were positively associated with the number of COVID-19 cases in the destination cities. They also found that the distance of a city from Wuhan is negatively associated with number of cases (Zhang et al., 2020). This study highlights how the COVID-19 virus can move efficiently across a country via its transportation systems.

As nations prepare for future waves of the COVID-19, airports that serve international travel need to develop more effective ways of identifying persons infected with COVID-19, whether exhibiting symptoms or not (CDC, 2020b; Lake, 2020). Better coordination with local public health officials in jurisdictions with major airports is also warranted. At the international level every airport, or governments with airports in their jurisdiction, should implement strict countermeasures (Nakamura and Managi, 2020). These communities need to be better equipped to isolate travelers, airport employees and nearby residents to effectively contain infections. Efforts need to be made to protect airline and airport employees. Employees need proper equipment to protect themselves from being infected. Their workspaces need to be modified to facilitate social distancing and procedures for properly cleaning facilities need to be implemented.

Fever screening at airports is the primary means of identifying infected travelers (Cho and Yoon, 2014). However, screening travelers for fever is not adequate to detect COVID-19 cases (Normile, 2020). A simulation study estimated that 46 percent of infected travelers would not be detected by fever screening techniques (Quilty et al., 2020). Reverse transcription polymerase change reaction (RT-PCR) is a reliable test in detecting both symptomatic and asymptomatic COVID-19 cases (Bwire and Paulo, 2020). However, the real time RT-PCR may not be practical for air travelers because it take the average laboratory 6–8 h to process the test samples (Jawerth, 2020).

As international travel begins to resume, some countries (Lieberman, 2020; TheNational, 2020) have required passengers to show a negative COVID-19 test that was performed recently, usually in the past 72 h, prior to admission at their ports of entry. Other countries have started offering COVID-19 testing upon arrival at the airport (Lieberman, 2020). Some airlines have responded (Craig, 2020; Lieberman, 2020) by requiring passengers to present a recent negative COVID-19 test prior to boarding their aircraft. Which strategy is most effective in preventing the entry and spread of COVID-19 remains to be seen. Given the demonstrated role of airports in the spread of COVID-19, strategies such as these may prove to be crucial in the prevention of COVID-19 transmission. Developing standards when caring for and transporting patients with suspected or confirmed infection with SARS-CoV-2 (AMPA, 2020), and specific training for the air ambulance aircrew (Martin, 2020) are other strategies to be considered. Using these strategies in domestic airports may also be worth considering especially for larger countries who wish to contain outbreaks in certain domestic regions.

While the International Air Transport Association (IATA) has released guidelines on the criteria for the use of COVID-19 testing in the air travel process (IATA, 2020), these are primarily non-binding recommendations. Apart from the logistics of screening, testing, and follow-up of passengers, another key issue is the costs involved in testing. The IATA has supported the World Health Organization's (WHO) recommendation that governments bear the cost for mandatory testing or offer tests at cost price for voluntary testing. Whether governments adopt this recommendation remains to be seen. Formal treaties or agreements and their enforcement might be necessary for testing to become more widespread at airports and other ports of entry. Alternatively, countries may pool resources to provide funding for COVID-19 testing among travelers (IATA, 2020).

In addition to improved methods of identifying cases, advanced methods of contact tracing must be deployed. South Korea is at the forefront of this exploring the use of smartphone tracking for new airport arrivals (Josh Smith, 2020; Mangan, 2020). However, using technology to track and identify persons and infected travelers who were in contact with infected individuals raises privacy concerns and may be a violation of individuals’ civil liberties. It is not only an invasion of the privacy of infected persons but also those persons who came in contact with them and did not consent to be surveilled.

As reported in several studies (Yancy, 2020; Price-Haywood et al., 2020, Laurencin and Mcclinton, 2020) community racial composition was associated higher rates of COVID cases and death. We also find race and ethnic disparities in COVID cases and deaths. Counties with higher percentage of Black Non-Hispanic, Hispanics and Native Americans were at greater risk for COVID cases and deaths. These race and ethnic disparities are exacerbated by the impact of housing density because minority households tend to be larger than White Non-Hispanic households.

Conclusions

Using descriptive, bivariate, and multivariate analyses, we explored the associations between COVID-19 cases and death rates and proximity to and size of airports, train stations, and public transportation. This study highlights the need for robust policies and procedures to detected infected travelers and conduct effective contact tracing to prevent the spread of the future waves of the COVID-19. Otherwise, transportation workers and residents of communities near airports and transportation hubs will be risk for future outbreaks. Future studies should look at patient level data to assess this association between air travelers, transportation employees and exposure the COVID-19.

Funding

This research is not funded by any organization.

CRediT authorship contribution statement

Darrell J. Gaskin: Conceptualization, Funding acquisition, Methodology, Validation, Writing - original draft, Writing - review & editing. Hossein Zare: Conceptualization, Data curation, Formal analysis, Methodology, Software, Validation, Writing - original draft, Writing - review & editing. Benjo A. Delarmente: Methodology, Validation, Writing - original draft, Writing - review & editing.

Declaration of competing interest

The Author(s) declare(s) that there is no conflict of interest.

	Negative Binomial Regression				Cox Regression
	Number of COVID-19Cases in County		Number of COVID-19Deaths in County		Time to First COVID-19Case in County		Time to First COVID-19Death in County
	IRR	CI-95	IRR	CI-95	HR	CI-95	HR	CI-95
Distance to airport: Ref: >50 miles
25> Miles to airport	1.367***	[1.164]-[1.606]	1.511***	[1.215]-[1.880]	1.610***	[1.409]-[1.839]	1.328***	[1.151]-[1.531]
25-50 Miles to airport	1.215***	[1.097]-[1.346]	1.216*	[1.012]-[1.460]	1.284***	[1.160]-[1.421]	1.154***	[1.062]-[1.254]
Num. of airport	1.011	[0.994]-[1.028]	1.005	[0.971]-[1.040]	1.013	[0.992]-[1.033]	1.027**	[1.008]-[1.047]
Volume of AirportRef: No volume
Medium volume	1.215*	[1.005]-[1.469]	1.233	[0.904]-[1.681]	1.354**	[1.109]-[1.654]	1.083	[0.842]-[1.393]
High volume	0.316***	[0.196]-[0.511]	0.243***	[0.137]-[0.430]	0.380***	[0.229]-[0.631]	0.751	[0.345]-[1.633]
Num. of train stations	1.096	[0.960]-[1.252]	1.266**	[1.059]-[1.512]	1.335***	[1.174]-[1.518]	1.235**	[1.067]-[1.429]
Being in counties or nearby counties with major 13 Airports	0.537**	[0.339]-[0.851]	0.467*	[0.248]-[0.877]	1.336	[0.637]-[2.803]	0.860	[0.384]-[1.924]
Being an international airport	1.108	[0.905]-[1.357]	1.128	[0.875]-[1.454]	1.516***	[1.203]-[1.911]	1.487**	[1.171]-[1.889]
CommuteRef: Work from home
Drove alone	1.030***	[1.024]-[1.035]	1.032***	[1.024]-[1.039]	1.012***	[1.007]-[1.017]	1.008***	[1.004]-[1.012]
Carpooled	1.012	[0.994]-[1.029]	0.987	[0.958]-[1.017]	0.986	[0.971]-[1.001]	0.989	[0.972]-[1.006]
Public trans	1.040*	[1.007]-[1.075]	1.090***	[1.038]-[1.144]	1.081***	[1.042]-[1.121]	1.084***	[1.052]-[1.116]
Walked	0.899***	[0.864]-[0.935]	0.890***	[0.852]-[0.930]	0.962**	[0.936]-[0.988]	0.938**	[0.902]-[0.976]
Other	1.113***	[1.049]-[1.180]	1.052	[0.976]-[1.134]	1.018	[0.973]-[1.065]	1.013	[0.968]-[1.060]
Commute Traveling time	0.996	[0.977]-[1.015]	1.018	[0.991]-[1.045]	1.021**	[1.007]-[1.036]	1.012	[0.998]-[1.026]
County Population
Total county population (/1000)	1.004***	[1.003]-[1.006]	1.005***	[1.003]-[1.006]	1.002***	[1.001]-[1.002]	1.001***	[1.001]-[1.001]
Population density (/100)	0.998***	[0.997]-[0.999]	0.997***	[0.996]-[0.998]	1.001***	[1.001]-[1.002]	1.001***	[1.001]-[1.002]
N	3012		3012		2968		2968

*p < 0.05, **p < 0.01, ***p < 0.001.

Sources: JHU-COVID-19 Data (Updated August 21, 2020), Airport and Amtrak Station Data (2019), ACS Data (2018)

Notes:

1) All models have been controlled by race/ethnicity groups, population density, number of people at household, age categories and percent of poverty.

2) We have clustered errors by state to capture states policies and interventions.

3) New York and California excluded from the analysis.

	Negative Binomial Regression				Cox Regression
	Number of COVID-19Cases in County		Number of COVID-19Deaths in County		Time to First COVID-19Cases in County		Time to First COVID-19Death in County
	IRR	CI-95	IRR	CI-95	HR	CI-95	HR	CI-95
Distance to airport in miles	0.998***	[0.998]-[0.999]	0.998***	[0.996]-[0.999]	0.997***	[0.996]-[0.998]	0.999***	[0.998]-[0.999]
Number of airport	1.029***	[1.016]-[1.043]	1.034**	[1.013]-[1.056]	1.014	[0.994]-[1.034]	1.036***	[1.024]-[1.049]
Size of airport (volume passenger/1,000,000)	0.994	[0.975]-[1.014]	0.983	[0.962]-[1.005]	0.981	[0.952]-[1.011]	0.988	[0.949]-[1.028]
Num. of train stations	0.967	[0.844]-[1.109]	1.017	[0.797]-[1.299]	1.058	[0.949]-[1.179]	0.996	[0.944]-[1.051]
Being in counties or nearby counties with major 13 Airports	0.521*	[0.307]-[0.884]	0.592	[0.275]-[1.275]	1.252	[0.623]-[2.515]	1.073	[0.640]-[1.801]
Being an international airport	1.207	[0.948]-[1.536]	1.176	[0.890]-[1.554]	2.066***	[1.637]-[2.607]	1.631***	[1.264]-[2.105]
CommuteRef: Work from home
Drove alone	1.031***	[1.026]-[1.037]	1.035***	[1.027]-[1.042]	1.013***	[1.008]-[1.018]	1.009***	[1.005]-[1.013]
Carpooled	1.005	[0.985]-[1.025]	0.979	[0.948]-[1.010]	0.986	[0.971]-[1.003]	0.988	[0.972]-[1.004]
Public trans	1.058*	[1.011]-[1.107]	1.107***	[1.049]-[1.169]	1.086***	[1.041]-[1.132]	1.108***	[1.081]-[1.136]
Walked	0.902***	[0.870]-[0.935]	0.897***	[0.858]-[0.938]	0.972	[0.943]-[1.002]	0.935***	[0.899]-[0.973]
Other	1.108***	[1.047]-[1.172]	1.051	[0.973]-[1.135]	1.019	[0.975]-[1.065]	1.015	[0.970]-[1.062]
Commute Traveling time	0.994	[0.976]-[1.012]	1.011	[0.987]-[1.036]	1.016*	[1.002]-[1.030]	1.009	[0.995]-[1.023]
County Population
Total county population (/1000)	1.003***	[1.002]-[1.005]	1.004***	[1.002]-[1.005]	1.001***	[1.001]-[1.002]	1.001***	[1.000]-[1.001]
Population density (/100)	0.999	[0.998]-[1.001]	0.998	[0.997]-[1.000]	1.001***	[1.001]-[1.002]	1.001***	[1.001]-[1.002]
N	3132		3132		3084		3084

*p < 0.05, **p < 0.01, ***p < 0.001.

Sources: JHU-COVID-19 Data (Updated August 21, 2020), Airport and Amtrak Station Data (2019), ACS Data (2018)

Notes:

1) All models have been controlled by race/ethnicity groups, population density, number of people at household, age categories and percent of poverty.

2) We have clustered errors by state to capture states policies and interventions.

	Negative Binomial Regression
	Number of COVID-19Cases in County		Number of COVID-19Deaths in County
	IRR	CI-95	IRR	CI-95
Distance to airport: Ref: >50 miles
25> Miles to airport	1.416**	[1.108]-[1.810]	1.657***	[1.273]-[2.156]
25-50 Miles to airport	1.210	[0.994]-[1.472]	1.210	[0.985]-[1.486]
Num. of airport	0.991	[0.957]-[1.025]	0.981	[0.943]-[1.021]
Volume of AirportRef: No volume
Medium volume	1.779**	[1.246]-[2.542]	1.456	[0.985]-[2.152]
High volume	0.562	[0.261]-[1.208]	0.358*	[0.135]-[0.954]
Num. of train stations	0.985	[0.783]-[1.239]	1.050	[0.740]-[1.491]
Being in counties or nearby counties with major 13 Airports	0.477	[0.223]-[1.020]	0.462	[0.165]-[1.290]
Being an international airport	1.197	[0.852]-[1.683]	1.354	[0.902]-[2.032]
CommuteRef: Work from home
Drove alone	1.024***	[1.016]-[1.033]	1.029***	[1.019]-[1.040]
Carpooled	1.020	[0.979]-[1.063]	0.986	[0.943]-[1.032]
Public trans	1.155***	[1.063]-[1.255]	1.144**	[1.043]-[1.255]
Walked	0.917**	[0.869]-[0.969]	0.952	[0.892]-[1.016]
Other	1.011	[0.925]-[1.105]	1.051	[0.948]-[1.165]
Commute Traveling time	0.994	[0.968]-[1.022]	1.005	[0.970]-[1.043]
County Population
Total county population (/1000)	1.003***	[1.002]-[1.005]	1.004***	[1.002]-[1.006]
Population density (/100)	0.998	[0.996]-[1.000]	0.997*	[0.995]-[1.000]
N	2891		2891

*p < 0.05, **p < 0.01, ***p < 0.001.

Sources: JHU-COVID-19 Data (Updated May 5, 2020), Airport and Amtrak Station Data (2019), ACS Data (2018)

Notes:

1) All models have been controlled by race/ethnicity groups, population, population density, number of people at household, age categories and percent of poverty.

2) We have clustered errors by state to capture states policies and interventions.