key: cord-0621608-q41m5kzh
authors: Sun, Qianqian; Pan, Yixuan; Zhou, Weiyi; Xiong, Chenfeng; Zhang, Lei
title: Quantifying the influence of inter-county mobility patterns on the COVID-19 outbreak in the United States
date: 2020-06-24
journal: nan
DOI: nan
sha: 63a35f8f29620a4007a2f79056c61a1b78f2b015
doc_id: 621608
cord_uid: q41m5kzh

As a highly infectious respiratory disease, COVID-19 has become a pandemic that threatens global health. Without an effective treatment, non-pharmaceutical interventions, such as travel restrictions, have been widely promoted to mitigate the outbreak. Current studies analyze mobility metrics such as travel distance; however, there is a lack of research on interzonal travel flow and its impact on the pandemic. Our study specifically focuses on the inter-county mobility pattern and its influence on the COVID-19 spread in the United States. To retrieve real-world mobility patterns, we utilize an integrated set of mobile device location data including over 100 million anonymous devices. We first investigate the nationwide temporal trend and spatial distribution of inter-county mobility. Then we zoom in on the epicenter of the U.S. outbreak, New York City, and evaluate the impacts of its outflow on other counties. Finally, we develop a"log-linear double-risk"model at the county level to quantify the influence of both"external risk"imported by inter-county mobility flows and the"internal risk"defined as the vulnerability of a county in terms of population with high-risk phenotypes. Our study enhances the situation awareness of inter-county mobility in the U.S. and can help improve non-pharmaceutical interventions for COVID-19.

control measures target restricting travel to mitigate the outbreak 1, 4-8 . Previous studies found that restrictions on long-distance travel are effective in the early stage while restrictions on local travel become more important after the virus disseminates [9] [10] [11] [12] . However, there lacks research on interzonal population flow at a large scale, especially within the United States. In this study, we examine the inter-county mobility patterns under various control measures (declaration of a national emergency, stay-at-home order, partial reopening), which include both local and longdistance travel based on real-world observations. We also specifically investigate New York City, which saw the largest outbreak in the United States, from two aspects: its role in the intercounty mobility and the impact of outflow from NYC on the outbreak severity in those destination counties. Moreover, we quantify the impacts of inter-county travel on the spread of COVID-19 in the United States using a "log-linear double-risk" model. For each county, the model considers both the external risks (sum of inflow from infected areas weighted by the outbreak severity in origin counties), and the internal risks (the vulnerability based on high-risk phenotypes [13] [14] [15] including people age 65 and over, male, African Americans, and low-income households). Mobile device location data is known for its capability of capturing timely and realworld trajectories at a large scale. The population flow estimates from mobile device location data are believed valuable to help curb the pandemic [16] [17] [18] . We thus consider mobile device location data as an appropriate data source to estimate the inter-county mobility patterns. To retrieve the real-world and real-time mobility patterns, we utilize an integrated set of mobile device location data, which involves over 100 million anonymous devices at a monthly basis.

Based on our previously developed algorithms 19 , we have estimated daily inter-county trip tables between 3,143 counties and county-equivalents from January 1, 2020 to May 15, 2020 (weekends and holidays are removed from analysis).

We first presented the temporal dynamics of nationwide inter-county trips in Fig 1a, where the total number of inflow inter-county trips of all the counties is examined. Based on the national trend, we recognized four stages: pre-pandemic (from January 1 to March 13), behavior change (from March 14 to April 13), quarantine fatigue (from April 14 to April 23), and partial reopening (after April 23). The pre-pandemic stage experienced a stable level before February 14 and a slight upward trend between February 15 and March 13, when the national emergency was declared. According to monthly variations of traffic volume, there was supposed to be an upward trend from January to March 20 . During the behavior change stage, the inter-county trip volume rapidly decreased within three weeks and then bottomed out on April 13, when the total intercounty trips decreased by 35%. After that, the inter-county trip total began to bounce back during the quarantine fatigue stage, which accompanied a consistent increase of confirmed COVID-19 cases. We defined this stage as quarantine fatigue because of an obvious rebound despite ongoing nationwide travel restrictions. On the week of April 24, states began deploying local phase-by-phase reopening plans by partially reopening selected businesses. During this stage, the inter-county trip volume kept rising even though the increase in confirmed cases had not slowed down.

We further evaluated the spatial differences between counties regarding the changes of total inflow inter-county trips and such trips specifically from NYC (Fig. 1b, c, d, e) . We calculated the weekday average from January 2 to January 31 excluding holidays as the baseline (Fig. 1b) .

Then we computed the weekday average week after week and the percentage change compared with the baseline. The week of March 9 presented widely increase instead of reduction (Fig. 1c) while the week of April 6 showed the most reduction (Fig. 1d) . Overall, eastern counties present a more significant reduction than western counties. For the week of the national emergency declaration, 83% of counties still showed a percentage increase in the in-flow inter-county trips while the five counties in NYC show a slight reduction of inflow trips ranging from -5% to 0 ( Fig. 1c) . It may imply that travelers were cancelling trips to NYC or relocating themselves to places with lower infection risk. Compared with the daily average number of destination counties in January, 554 (17% of all counties), the inter-county trips from NYC still had a widespread distribution ending at 518 destination counties (16% of all counties). During the week of April 6, 92% of counties experienced a percentage reduction in in-flow inter-county trips, and the spatial distribution of inter-county trips from NYC was clearly narrowed down to 272 (8%) destination counties. During the most recent week, 48% of counties show a percentage increase in inflow trips, especially in western counties (Fig. 1e) .

After evaluating the inflow inter-county volume at each county, we specifically investigated the greater New York City (NYC) area, consisting of New York County, Bronx County, Queens County, Kings County, and Richmond County, due to its intensive travel interactions with other counties. As the largest transportation hub and the most densely populated city in the United States, NYC became the U.S. epicenter at an early stage. We evaluated four metrics related to inter-county trips for NYC: total inflow trips, total outflow trips, the number of origin counties accounting for those inflow trips, and the number of destination counties accounting for those outflow trips (Extended Fig. 1 ). We found that NYC stayed in the top three even after national emergency in all those four aspects when compared with other counties. This raised concerns on inter-county disease transmissions.

From January 1 to May 15, the inter-county trips from NYC directly distributed to 42% counties in the United States, including Alaska and Hawaii, and might have a wider impact considering the final destinations of those trips. Despite that the inter-county trips generated from NYC have been reduced by 60% at the most, the overall volume is still high compared with other counties.

We first evaluated the influence of such trips on local outbreaks by zooming in on the top twelve destination counties (Fig. 2 ). Fig. 2b shows the temporal changes of inter-county trips coming from NYC to the top twelve destination counties, and Fig. 2a shows the temporal changes in the cumulative cases in those counties. Nassau County and Westchester County in New York are the two counties with most inter-county trips from NYC before the pandemic. Therefore, they also have the most cumulative cases during the first three weeks after the pandemic began. As the confirmed cases increased, Nassau County and Westchester County experienced a sharp decrease in the trip volume from NYC. In the meantime, the decrease in trips to Suffolk County, NY was not as significant, which later made Suffolk County succeed Nassau County as the county with the second-most confirmed cases. Although Hudson ranked third before pandemic, it has experienced the maximum percentage reduction in the inter-county trips from NYC and thus had a lower rank in confirmed cases.

As the influence of inter-county trips from NYC was unveiled, we further investigated the impacts by examining the day-by-day correlations between such trips and the cumulative COVID-19 cases per capita (per thousand people) of the 1360 destination counties. Since COVID-19 has an incubation period, we have considered four scenarios: no time lag, one-week lag, two-week lag, and three-week lag (Fig. 3) . For example, the one-week lag scenario calculates the correlation between the inter-county trip volume from NYC and the cumulative case number per capita one week later. Overall, there is a significant positive correlation between inter-county trips from NYC and the outbreak severity in the destination county. The correlation became more significant as the outbreak kept spreading. And it can be as high as 0.68 (Pearson's r) on Apr 23 in the three-week lag scenario and 0.72 (Spearman's rs) on Apr 22 in the threeweek lag scenario. Among the four scenarios, the ones with time lag show a stronger correlation than that without time lag, and the correlation strength is similar between the ones with time lag.

We thus conclude that travel from severely infected areas would significantly contribute to local outbreaks with time lags.

Quantifying the influence of the inter-county trips via "log-linear double-risk" models

Following the case study in NYC, we expanded our study to quantifying the general influence of inter-county trips from all the infected counties. We innovatively developed a "log-linear double- is the constant in the model. We fit the model into twenty scenarios as shown in Fig. 4 . The results show that the model with one-week lag outperforms the others when applied to the whole after-pandemic announcement period (the combination of behavior change, quarantine fatigue, and partial reopening) with R 2 =0.62. For individual stages, the models with shorter time lags have a better fit, which may imply that the community transmissions contribute more to the outbreak severity than the imported cases from other counties. Another possible reason is that most inter-county trips are conducted by residents regularly, e.g., daily, which reduces the significance of an incubation period. Besides, the models fit the data better in the later stages. It may indicate that our assumption for the external risks, where the infected ratio of inter-county travelers is the same as that of the entire population in the origin county, became more literal along with the spread of COVID- 19 . Moreover, the model shows that the logged external risk is much more important than internal risk by improving R2 significantly (Extended Table 1 ).

This study focuses on the population flow between all counties in the U.S. and inspects how inter-county trips aggravate COVID-19 to provide insights into the epidemiological situation in 

Population flow drives spatio-temporal distribution of COVID-19 in China

The effect of human mobility and control measures on the COVID-19 epidemic in China

Isolation, Quarantine, Social Distancing and Community Containment: Pivotal Role for Old-Style Public Health Measures in the Novel Coronavirus (2019-nCoV) Outbreak

Characteristics of and Important Lessons from the Coronavirus Disease 2019 (COVID-19) Outbreak in China: Summary of a Report of 72 314 Cases From the Chinese Center for Disease Control and Prevention

First Cases of Coronavirus Disease

Taking the right measures to control covid-19. The Lancet Infectious Diseases

Coronavirus disease 2019 (covid-19): a perspective from china

An investigation of transmission control measures during the first 50 days of the COVID-19 epidemic in China

Disparities in the Population at Risk of Severe Illness From COVID-19 by Race/Ethnicity and Income. American journal of preventive medicine

COVID-19)-United States

COVID-19 and African Americans

Aggregated mobility data could help fight COVID-19

Mobile phone data for informing public health actions across the COVID-19 pandemic life cycle

Commentary: Containing the Ebola Outbreak -the Potential and Challenge of Mobile Network Data

An interactive COVID-19 mobility impact and social distancing analysis platform

Vehicle Volume Distributions by Classification

Human mobility trends during the COVID-19 pandemic in the United States

Staying at Home: Mobility Effects of COVID-19

Distribution of the COVID-19 epidemic and correlation with population emigration from Wuhan

Epidemiology and transmission of COVID-19 in 391 cases and 1286 of their close contacts in Shenzhen, China: a retrospective cohort study

The Incubation Period of Coronavirus Disease 2019 (COVID-19) From Publicly Reported Confirmed Cases: Estimation and Application

Early Transmission Dynamics in Wuhan, China, of Novel Coronavirus-Infected Pneumonia

We would like to thank and acknowledge our partners and data sources in this effort: (1) partial financial support from the U.S. Department of Transportation's Bureau of Transportation Statistics and the National Science Foundation's RAPID Program; (2) Amazon Web Service and its Senior Solutions Architect, Jianjun Xu, for providing cloud computing and technical support; 

LZ and QS designed the study. QS, YP, and WZ analyzed the data. QS, YP, and WZ interpreted the data. QS, YP, and WZ wrote the manuscript. All authors contributed to the final draft.

All authors declare no competing interests.

The aggregated inter-county trips will be published on a public site at data.covid.umd.edu. The authors declare no competing financial interests. Correspondence and requests for materials should be addressed to QS (qsun12@umd.edu). Reprints and permissions information is available at www.nature.com/reprints.