The interplay of spatial spread of COVID-19 and human mobility in the urban system of China during the Chinese New Year

Abstract

The rapid spread of infectious diseases is devastating to the healthcare systems of all countries. The dynamics of the spatial spread of epidemic have received considerable scientific attention. However, the understanding of the spatial variation of epidemic severity in the urban system is lagging. Using synchronized epidemic data and human mobility data, integrated with other multiple-sourced data, this study examines the interplay between disease spread of coronavirus disease (COVID-19) and inter-city and intra-city mobility among 319 Chinese cities. The results show a disease spreading process consisting of a major transfer (inter-city) diffusion before the Chinese New Year and a subsequent local (intra-city) diffusion after the Chinese New Year in the urban system of China. The variations in disease incidence between cities are mainly driven by inter-city mobility from Wuhan, the epidemic center of COVID-19. Cities that are closer to the epidemic center and with more population in the urban area will face higher risks of disease incidence. Warm and humid weather could help mitigate the spread of COVID-19. The extensive inter-city and intra-city travel interventions in China have reduced approximately 70% and 40% inter-city and intra-city mobility, respectively, and effectively slowed down the spread of the disease by minimizing human to human transmission together with other disease monitoring, control, and preventive measures. These findings could provide valuable insights into understanding the dynamics of disease spread in the urban system and help to respond to another new wave of pandemic in China and other parts of the world.

Keywords

Spatial spread infectious disease human mobility COVID-19 urban system

Introduction

Since December 2019, a highly contagious novel coronavirus disease (COVID-19) has been spreading within and beyond China. The first case in China was identified in Wuhan city in early December (Huang et al., 2020). The situation became severe after 10 January 2020 when the chunyun, the travel season during the Chinese New Year (CNY), began. Chunyun is regarded as the largest annual human migration in the world. Most people will go back to their hometowns and celebrate the traditional festival with their families. The travel season usually starts 15 days ahead of CNY and ends 25 days after. In 2019, there were nearly three billion journeys during the 40-day chunyun—almost every Chinese traveled more than twice on average. Wuhan is a highly populated city in central China and had 11.08 million usual residents in 2018. To contain the outbreak, Wuhan was quarantined at 10:00 am on 23 January 2020 restricting people from travelling out of Wuhan. Most of the cities in Hubei province also imposed the quarantine within the next two days. However, according to the official information released at the press conference on 26 January, approximately five million people had left Wuhan before the quarantine.

The movement of population could shape the spatio-temporal patterns of global pandemic (Balcan et al., 2009). Human travels played an important role in the disease spread (through human to human transmission), such as severe acute respiratory syndrome (SARS), Middle East Respiratory Syndrome, and influenza (Abdullah et al., 2004; Bajardi et al., 2011; Colizza et al., 2006, 2007; Findlater and Bogoch, 2018; Russell et al., 2008). The spatial spread of infectious diseases has received extensive scientific attention. Many efforts have been initiated to incorporate human mobility in epidemic modeling (Belik et al., 2011). As infectious diseases are transmitting from human to human, human movements such as international migration (Bedford et al., 2010; Quinn, 1994), commuting flows (Dalziel et al., 2013; Massaro et al., 2019; Viboud et al., 2006), travels via airline transportation (Colizza et al., 2006; Findlater and Bogoch, 2018; Gardner and Sarkar, 2013; Merler and Ajelli, 2010), and daily activities (Eubank et al., 2004), could contribute to the spread of diseases. Most of the earlier studies used aviation network, a relatively coarse human mobility dataset, to predict epidemic spread. Hufnagel et al. (2004) showed that incorporating a stochastic dispersal of individuals in a worldwide network, the probabilistic model could well predict the global spreading of SARS cases in 2003. The inclusion of air flow data was also crucial in predicting spatio-temporal evolution of H5N1 influenza (Colizza et al., 2007). Recently, some theoretic epidemic models were proposed to utilize mobility process down to the individual level (Belik et al., 2011; Meloni et al., 2011), despite the limited data availability of fine-grained mobility (Balcan et al., 2009). Using individual commuting data, a cross-city comparison study suggested heterogeneity of commuting patterns can explain the inter-city variation in epidemic dynamics (Dalziel et al., 2013).

As technology advanced, large-scale mobile phone data became available to study the spreading mechanism of infectious diseases. The fine-grained mobility data can improve prediction accuracy (Bengtsson et al., 2015), and real-time mobility data can help us better understand the transmission process and the effectiveness of different control measures on the epidemic spread (Kraemer et al., 2020). As the novel coronavirus became a pandemic, many companies, e.g. Google, Tencent, and Baidu, have generated aggregated mobility data that could help fight COVID-19. The research and public health response communities used these mobility data to study and refine interventions (Buckee et al., 2020). Kraemer et al. (2020), Lai et al. (2020), Tian et al. (2020), Chinazzi et al. (2020), Jia et al. (2020) and many studies have found the drastic control measures over human mobility in China substantially mitigated the spread of COVID-19.

Despite abundant studies on the spatial spread of infectious diseases, most of the studies focus on modeling at a certain scale, ranged from intra-city to international level, and few have integrated multiple perspectives. The importance of human mobility compared with other factors is rarely studied. Cold and dry weather is considered as the key driver of the pronounced seasonality of some infectious disease like inﬂuenza (Dalziel et al., 2018; Shaman and Kohn, 2009). The larger population size, larger house size, higher population density, and higher urbanization level often bring higher possibilities of contacts between people, thus bring more risks of disease spread to cities (Chowell et al., 2008; Dalziel et al., 2018; Grenfell and Bolker, 1998; Merler and Ajelli, 2010; Tian et al., 2018). From another perspective, higher urbanization level, higher GDP per capita, and better medical care resource (e.g., hospital beds per 1,000 people) often correlate with better living conditions and more timely treatment, which could mitigate the spread of certain diseases (Gardner et al., 2018; Meng et al., 2005). Moreover, the spatial scale matters. Whether these factors are effective at the city level is still unknown. Using synchronized epidemic data and human mobility big data, integrated with other multiple-sourced data, this study examines the spatial spread and dynamics of the novel COVID-19 among Chinese cities before and after the CNY in 2020. Three questions are examined: (1) How important is inter-city mobility in explaining the variations of disease incidence in the urban system? (2) What is the relationship between intra-city mobility and epidemic severity among cities? (3) Does human mobility still play an essential role in the disease spread after controlling for other critical factors?

Methodology and data

Study units

This study covers 320 administrative cities and units in mainland China in total, including 4 province-level cities, 272 prefecture-level cities, 26 prefecture-level units, and 18 county-level units, which are directly under the administration of provinces (see Figure S1 in the Supplemental Material). These administrative cities and units are independent of each other (no unit is under the administration of another). A detailed introduction to the administrative system of Chinese cities can be referred to Ma (2005). We refer to all these units as “cities” in this study.

Data

The epidemic data, human mobility data, weather data, and socioeconomic data were collected at the city level from multiple sources.

Epidemic data

We collected epidemic data mainly from two sources. (a) The daily aggregated data were obtained from the website of the National Health Commission of China (NHCC) (http://www.nhc.gov.cn). The national aggregated number as of 24:00 of the day would be reported on the second day. (b) The infected cases of each city were obtained from Ding Xiang Yuan (DXY, https://ncov.dxy.cn/ncovh5/view/pneumonia), a professional platform in the medical field. The website collects authoritative public information from local departments and provides a timely update.

Mobility data

We collected daily human mobility data from Baidu Mobility (http://qianxi.baidu.com) and Baidu Traffic (https://jiaotong.baidu.com/). Baidu provides location-based services to over 500,000 applications on mobile phones and responses to nearly 120 billion service requests every day (http://lbsyun.baidu.com/products/products/location). China has 847 million mobile Internet users by 2019, accounting for over 60% of the total population (China Internet Network Information Center, 2019). The large base of mobile internet users and the wide use of Baidu’s location services make the mobility data representative. Baidu provides daily inter-city and intra-city mobility data for 2020 and 2019 at the city level. Inter-city mobility index (IMI) refers to the relative inter-city travel volume. Intra-city travel intensity (ITI) measures the relative ratio between the number of intra-city trips and the number of residents in the city. We collected the mobility data from 8 January to 11 February in 2020 and the corresponding period in 2019. A more detailed description of mobility data could be found in Supplemental Material.

Weather data

We obtained the daily temperature and daily relative humidity data for each city from 20 January to 10 February 2020 from a Chinese weather website (http://www.xaoyo.com). The average temperature and absolute humidity were used in the analyses. The absolute humidity was calculated based on the temperature and relative humidity using the R package “humidity.”

Socioeconomic and population data

We obtained the socioeconomic and population data of each city from multiple sources: (a) China City Statistical Yearbook 2018 (NBSC, 2019); (b) Provincial Statistical Yearbooks 2018; (c) Local statistical report in 2018; (d) The 2015 One Percent Population Sample Survey of China (NBSC, 2015). In this study, the number of household registered population, administrative area, and hospital beds, GDP per capita of each city are obtained from (a) to (c). Considering that most of the Chinese people have returned to their hometown during the CNY, we used the household-registered population of each city in this study. Population density was calculated by dividing the household registered population size by the administrative area. Hospital beds per 1,000 people in each city was calculated by dividing the total number of hospital beds by the number of the registered population. The urbanization level and household size of each city were derived from (d). Urbanization level measures the percentage of the population living in the urban area of a city. It should be noted that usual residents (other than household registered population) are used in the definition of urbanization level. Household size refers to the average number of family members living together in a house.

Method

Study period denotation

This study mainly covers about one month around CNY, from 8 January to 11 February 2020. Week 1b and 2b refer to the first and second week before CNY (18–24 January and 11–17 January in 2020); week 1a and 2a refer to the first and second week after CNY (25–31 January and 1–7 February in 2020).

Definition of ratio of incidence

The ratio of incidence $R I (t, w)$ is defined as the ratio between the number of new cases during a given period from time $t + w$ to $t + 2 w$ and the number of new cases during the period from time $t$ to $t + w$ . We calculate $R I (t, w)$ using the formula

R I (t, w) = (C_{t + 2 w} - C_{t + w}) / (C_{t + w} - C_{t})

(1)

where

C_{t}

is the number of accumulative infected cases at time

t

;

w

is a period in the disease spread (here

w

represents one week).

R I (t, w)

can be used to evaluate the effect of measures taken during time

t

and

t + w

. If

R I (t, w)

is larger than 1, the disease is spreading to a larger scale. The speed of disease spread of Week 1a RI (1a, w) is measured by dividing the new cases during Week 2a by the new cases during Week 1a. The speed of disease spread of Week 2a RI (2a, w) is measured by dividing the new cases during Week 3a by the new cases during Week 2a.

Negative binomial regression

Negative binomial (NB) regression can better model over-dispersed count variables (Sparks et al., 2010). We performed NB regression to quantify the effect of human mobility and other factors on the epidemic severity in the urban system using the following equations

L n (E [Y]) = β_{0} + β_{1} L n (IMI) + β_{2} ITI + β_{3} L n (Pop) + β_{4} L n (Density) + β_{5} U L + β_{6} A H + L n (Exposure)

(2)

where the dependent variable

Y

is the number of infected cases during a period of time of each city. The explanatory variables are inter-city mobility index (IMI), intra-city travel intensity (ITI), population size (Pop), population density (Density), urbanization level (UL), and absolute humidity (AH) of each city. The exposure variable identifies the period of time in which the counts of infected cases are accumulated. For example, if the disease in a city was first reported on 25 January, then the number of days from the arrival of the disease to 3 February should be 10.

β_{0}, β_{1}, β_{2} \dots β_{6}

are regression coefficients to be estimated.

L n

is the natural logarithm. Some variables are highly skewed (e.g. IMI, population size, population density), and the range of values is very large. The log transformation could convert a variable to be more closed to a normal distribution and make the results less influenced by extreme values. Moreover, for some variables like IMI, we are more interested in the interpretation of how the percentage change of the relative mobility volume affects the disease incidence, which could be achieved by log-transformation.

Results

Two stages of the disease diffusion process

The process of spatial spread of COVID-19 mainly consisted of two stages, namely, transfer (inter-city) diffusion and local (intra-city) diffusion. Transfer diffusion refers to the process of disease spread that is predominantly driven by the high degree of human mobility between places (Smallman‐Raynor and Cliff, 2001). The epidemiological features show that the median incubation period is about 4 days (interquartile range 2–7) (Guan et al., 2020). The median time period from symptom onset to diagnosis has decreased from 12 days (range 8–18) in early January to 3 days (range 1–7) by early February (WHO, 2020). The potential patients who left Wuhan before the quarantine (23 January 2020) were likely to be diagnosed and reported at or before the beginning of February. The epidemic curve shows that the diagnosis of illness peaked around 3 February 2020 (Figure 1(a)), while the onset of illness peaked around 27 January (China CDC, 2020). A time lag of roughly 7 days is noticed between the onset and report date in late January and early February. As of 3 February, a total of 14,054 infected cases in 319 Chinese cities (excluding Wuhan) have been reported by the NHCC, which may primarily be the outcome of transfer diffusion.

Figure 1.

The epidemic and inter-city mobility during chunyun. (a) The epidemic curve of daily new cases in Chinese cities (excluding Wuhan) reported by NHCC. (b) Inter-city mobility index in China and Wuhan before and after CNY. (c) Flow index in traffic hubs after CNY. (d) Snapshots of the distribution of accumulative infected cases of COVID-19 in mainland China at four time points from 21 January to 11 February 2020. Grey indicates the population density, red indicates the infected individuals, and green indicates the discharged individuals.

The inter-city mobility pattern has been significantly changed because of COVID-19 (Figure 1(b)). The travel season is divided into two periods: the active period before CNY and the inactive period on and after CNY. IMI of China has decreased by 70% in the Two Weeks After CNY of 2020 compared with the same period of 2019. IMI from Wuhan to other cities dropped to close to 0 after CNY. A more detailed indicator, flow index, monitors the relative volume of visitors in traffic hubs every 30 min every day. After CNY, the daily peak volume of visitors in cities has reduced to 1/3 or 1/4 of the level compared with last year (Figure 1(c)).

Due to the limited inter-city mobility after CNY, the major process of transfer diffusion between cities may have completed during the active travel period before CNY. The majority of cases caused by transfer diffusion were likely to be reported on or before 3 February 2020 (onset on or before 27 January 2020). After that, the new cases in a city were expected to be contributed mainly by local diffusion. The snapshots of the distribution of accumulative infected cases provide the clues to the process of disease spread (Figure 1(d)). The epidemic was mainly limited to Wuhan city before CNY, but rapidly spread across the country after CNY. As of 11 February, a total of 25,095 infected cases in 321 cities (excluding Wuhan) have been reported.

Spatial variations of disease incidence and inter-city mobility

Spatial variations can be found in the disease incidence in the urban system (319 cities excluding Wuhan). There were 96, 86, 32, and 105 infected cities in eastern, central, northeastern, and western regions, respectively (See Regional Division in the Supplemental Material), accounting for 94%, 89%, 100%, and 78% of the total number of cities of each region. The average numbers of infected cases as of 3 February 2020 in the cities of the four regions were 33, 110, 8, and 12, respectively. We calculated the great-circle distance between each city and Wuhan using their longitudes and latitudes using R package “geosphere.” The Spearman Rank Correlation suggests that the number of accumulative cases as of 3 February 2020 is negatively correlated with the distance from Wuhan (Spearman’s $ρ$ = −0.62, $P < 0.001$ ), which suggests that cities closer to Wuhan were more severely affected.

The distribution of outflows from Wuhan in the Two Weeks Before CNY was uneven. A small proportion of cities absorbed the majority of the travel flows. For example, Xiaogan and Huanggang, two hardest-hit cities in Hubei province, received 27% of the outflows from Wuhan. The spatial pattern of inter-city travel flows shows that cities in Hubei province were the primary receivers of flows from Wuhan (Figure 2). Cities in the surrounding provinces (e.g. Henan, Hunan, and Anhui) and some super-large cities which may have close business and trade connections with Wuhan (e.g. Beijing, Guangzhou, Shenzhen, and Chongqing) were also important receivers. These cities face higher risks in the disease spread. Top 45 cities have received 85% of the travel flows from Wuhan and contributed to 74% of the infected cases as of 3 February. The volume of mobility flows from Wuhan is negatively correlated with the distance from Wuhan (Spearman’s $ρ$ = –0.75, $P < 0.001$ ), which means cities closer to Wuhan have received more travelers from Wuhan than cities farther away.

Figure 2.

Inter-city mobility flows from Wuhan and disease spread in the urban system. Node color indicates the relative volume of flows from Wuhan in the Two Weeks Before CNY. Line color indicates the destination of travel flows. Larger node size indicates more infected cases in the city as of 3 February 2020. Thicker lines represent larger volumes of flows from Wuhan. Top 45 cities absorbed 85% of the outflows from Wuhan.

A positive correlation between inter-city mobility and epidemic severity

The accumulative number of infected cases correlates closely with mobility flows from Wuhan (Figure 3(a)). First, we measured the strength of associations between mobility flows from Wuhan in the Two Weeks Before CNY and the accumulative number of infected cases in cities. The values of Spearman’s $ρ$ start from 0.59 on 24 January to a peak of 0.85 around 3 February (all associated $P values < 0.001$ ) (Figure 3(b)). The more people from the epidemic center a city receives, the more infected cases the city is likely to have. During 4–11 February, Spearman’s $ρ$ stays around 0.85 with small fluctuations, which suggests that the local diffusion may have maintained at a low level broadly. Otherwise, the accumulative number of infected cases and the mobility flows from Wuhan in the Two Weeks Before CNY could become less correlated.

Figure 3.

Inter-city mobility flows from Wuhan, infected cases, and the correlation. (a) The number of accumulative infected cases in each city as of 3 February 2020 by the percentage of mobility flows from Wuhan in the Two Weeks Before CNY. Spearman’s rank correlation is 0.85 (P < 0.001). Node color indicates the volume of travel flows the city received from Wuhan in the Two Weeks Before CNY. Node size indicates the number of infected cases as of 3 February 2020. (b) The correlation coefficient Spearman’s ρ between the accumulative infected cases and the relative volume of mobility flows from Wuhan in cities in the Two Weeks Before CNY.

Relationship between epidemic severity and intra-city mobility

COVID-19 also changed the pattern of intra-city mobility significantly. ITI usually decreases on weekends and increases on workdays, indicating its close correlation with commuting patterns (Figure 4(a)). In previous years, people usually begin to return to work in Week 2a. However, this year, the average ITI value in Week 2a (2.31, 95%CI: 2.23–2.39) is even lower than in Week 1a (3.03, 95%CI: 2.93–3.14) (Figure 4(b)). The average level in the Two Weeks After CNY in this year (2.67, 95%CI: 2.58–2.76) has decreased by over 40% compared with the same period of last year (4.60, 95%CI: 4.50–4.70).

Figure 4.

Intra-city mobility and the relationship with epidemic size. (a) Intra-city travel intensity (ITI) before and after the CNY in 2020 and 2019. (b) Number of cities with different ITI values in Week 1a and 1b of 2020 and 2019 (smoothed line). (c) Number of new cases (4–11 February 2020) in each city by average ITI values one week earlier (28 January to 4 February 2020). Cities with no new cases are not shown. Spearman’s rank correlation is –0.40 (P < 0.001). Node color indicates the average ITI value. Node size indicates the number of new infected cases. (d) ITI ratio between Week 1a and Week 1b in four categories of cities classified by epidemic size (as of 3 February 2020). Boxes represent the interquartile range of the ITI ratio. Horizontal lines and triangles indicate the median and mean values. (e) The ratio of incidence in 161 cities with relatively strong interventions. More cities have a lower speed of disease spread after minimizing the intra-city mobilities.

The relationship between epidemic severity and intra-city mobility is a complicated interaction. We assumed that the reported infections are related to the activities that happened in the city one week ago or earlier (considering the incubation period and the time from onset to diagnosis). First, we assessed the correlation between the number of cumulative cases as of 3 February and the average ITI values one week earlier (18–27 January) among 319 cities. No significant correlation is found (Spearman’s $ρ$ = 0.09, $P = 0.12$ ). Second, we assessed the correlation between the number of new cases during 4–11 February and the average ITI values one week earlier (28 January to 4 February) among 319 cities. The result revealed a negative correlation (Spearman’s $ρ$ = –0.40, $P < 0.001$ ). It seems that cities with larger epidemic sizes will have lower ITI values (Figure 4(c)). The possible explanation is that intra-city mobility in Chinese cities is profoundly affected by the disease occurrence. The spatial pattern of average ITI values shows that many severely infected cities in Hubei, Henan, Zhejiang, Hebei, Shandong, Jiangsu, and Guangdong maintained at a very low mobility level (see Figure S2 in the Supplemental Material). The more severe the outbreak, the more drastic strategies the government will adopt, and less intra-city mobilities people will have.

Variations of intra-city travel intensity

The degree of interventions varies in different cities. We classified cities into four categories corresponding to the quartiles of cumulative cases as of 3 February (from small to large, Q1 to Q4). The average numbers of infected cases in Q1–Q4 are 2, 7, 16, and 227, respectively. We plotted the ratio of ITI between Week 1a and Week 1b of each category (Figure 4(d)). The results show that ITI in cities with larger epidemic sizes decreases more significantly. The median ratios of ITI in Q3 (54%) and Q4 (46%) are significantly lower compared to Q1 (67%) and Q2 (65%) (Wilcoxon test, $P values < 0.001$ ). The difference in the ITI ratios between Q1 (67%) and Q2 (65%) is not statistically significant (Wilcoxon test, $P = 0.06$ ). The reason might be that the number of infected cases in Q1 and Q2 is relatively small ( $\leq 10$ ), and the situation is under control. So, people’s behavior in Q1 and Q2 are relatively less affected. However, compared with last year, all cities have significantly lower ITI ratios (Figure 4(d), Wilcoxon test, $P values < 0.001$ ).

Effects of intra-city travel restrictions

To test the effect of intra-city travel interventions on the disease spread, we further calculate the ratio of disease incidence $R I (t, w)$ (see Section “Method,” equation (1) for details) in 161 cities with relatively strong interventions (Q3 and Q4). The results show that with lower intra-city mobilities, more cities have lower speed of disease spread in Week 2a compared with Week 1a (Figure 4(e)). The average RI among 161 cities dropped from 1.26 (95%CI: 1.11–1.42) in Week 1a to 0.58 (95%CI: 0.48–0.68) in Week 2a. Although the decreasing ITI did not stop the growth of the absolute number of new cases, it effectively slowed the spread of the disease. However, from another perspective, even strict controls can hardly stop the spread. Once the epidemic ramps up, the endogenous force of disease transmission (e.g. exponential growth) will become the major driver (Balcan et al., 2009). Building early warning systems and taking measures at the emergence of the epidemic may be effective in containing the outbreak.

Role of human mobility and other factors

Does human mobility still play an essential role of disease spread in the urban system when other factors are considered in the analysis? To answer this question, first, we assessed the correlation between each pair of the following variables which may contribute to the disease spread in the urban system: IMI from Wuhan (inter-city mobility flows from Wuhan to each city in the Two Weeks Before CNY), ITI (average ITI value in the Two Weeks After CNY), distance from Wuhan, population size, population density, urbanization level, GDP per capita, hospital beds per 1,000 people, temperature, and absolute humidity (AH) (see Table S1 in the Supplemental Material). Strong correlations between IMI from Wuhan and distance from Wuhan (Spearman $ρ$ = –0.75, P < 0.001), urbanization level and GDP per capita (Spearman $ρ$ = 0.77, P <0.001), temperature and AH (Spearman $ρ$ = 0.95, P < 0.001) were found.

Second, we measured the Spearman correlations between the number of accumulative cases N (as of 3 February 2020), the total number of new cases $Δ$ N (during 4–11 February), and all variables. The mean and standard deviation of variables and the results of associations are listed in Table 1. Because of the incubation period and the time from onset to diagnosis, the time lag of one week was considered in testing the effect of intra-city mobility, e.g. average ITI value during 18–27 January 2020 was used to test the correlation with $N$ , average ITI value during 28 January to 4 February 2020 was used to test the association with $Δ N .$ Most of the variables are independently statistically correlated with the epidemic size except household size and hospital beds. The reason might be that China has adopted highest-level containment strategies on and after 23 January 2020, including the prompt detection and isolation of suspected cases. For most of the cities, the local diffusion maintained at a low level. So, these two factors which are expected to be associated with epidemic severity show no significance.

Table 1.

Descriptive statistics of variables and relationship between the accumulative infected cases $N$ , increased infected cases $Δ N$ , and potential factors at the city level given by Spearman’s correlation coefficient.

Variables	Mean	Std. Deviation	$ρ for N$ (Group 1)	$ρ for Δ N$ (Group 2)
IMI from Wuhan	3,250.80	13,290.71	0.85***	0.77***
ITI	2.67	0.80	0.09	–0.40***
Distance from Wuhan (km)	943.97	582.28	–0.62***	–0.54***
Population size (million)	4.21	3.23	0.56***	0.55***
Population density (persons/km²)	416.30	348.86	0.56***	0.52***
Urbanization level (%)	53.15	14.73	0.29***	0.27***
Household size (persons)	3.26	0.44	–0.07	–0.06
Hospital beds (per 1,000 persons)	4.60	1.73	0.004	0.03
GDP per capita (USD)	8,193.82	4,611.52	0.34***	0.27***
Temperature ( $°C$ )	4.32	8.69	0.25***	0.11
Absolute humidity (g/km³)	5.41	3.31	0.28***	0.16**
Number of observations	319	319	319	319

Note:

$N$ is the number of accumulative cases as of 3 February 2020. $Δ N$ is the total number of new cases during 4–11 February 2020. IMI from Wuhan refers to the IMI flows from Wuhan in the Two Weeks Before CNY (11–24 January 2020). ITI in Group 1 refers to the average ITI value during 18–27 January 2020. ITI in Group 2 refers to the average ITI value during 28 January to 4 February 2020. The exchange rate of the RMB against the USD is 6.75.

***P < 0.001, **P < 0.01, *P < 0.05.

CNY: Chinese New Year; IMI: inter-city mobility index; ITI: intra-city travel intensity.

Third, we performed NB regression to identify and quantify the effect of human mobility and other factors on the spread of COVID-19 in 319 Chinese cities (see Section “Method,” equation (2) for details). We dropped distance from Wuhan, GDP per capita, and temperature in the models to avoid collinearity. In Model 2, we replaced the variable IMI from Wuhan by the number of active cases (active cases = accumulative cases – deaths – discharged cases) as of 3 February 2020. For infectious diseases, the counts of past cases should have a great influence on the new counts of future cases. The variable IMI from Wuhan and active patients are closely correlated (spearman’s $ρ$ = 0.84, P < 0.001). However, employing the number of active cases can improve the quality of Model 2. The McFadden’s pseudo-R² increased from 0.168 (using IMI from Wuhan) to 0.213 (using active cases). More importantly, during 4–11 February, the intercity mobility was limited, particularly from Wuhan (Figure 2(b)). We assume that a local diffusion process contributes to the disease spread during this period. We are more interested in the role of intra-city mobility on the disease spread during this process. The results of the regression are presented in Table 2.

Table 2.

Associations between accumulative cases, new cases, and factors, evaluated by negative binomial regression models.

	Model 1 ( $N)$		Model 2 ( $Δ N$ )
Variables	Coeff.	95% CI	Coeff.	95% CI
(Intercept)	–3.91	(–4.75, –3.07)	–2.23	(–3.15, –1.32)
IMI from Wuhan (logarithm)	0.69***	(0.64, 0.74)	–	–
ITI	–0.08	(–0.22, 0.07)	–0.04	(–0.20, 0.12)
Population size (million, logarithm)	–0.005	(–0.12, 0.11)	0.03	(–0.12, 0.18)
Population density (persons/km², logarithm)	–0.05	(–0.16, 0.05)	0.06	(–0.07, 0.18)
Urbanization level (%)	0.01***	(0.005, 0.015)	0.002	(–0.009, 0.007)
Absolute humidity (g/km³)	0.002	(–0.02, 0.03)	–0.04*	(–0.07, –0.007)
Active cases (as of 3 February logarithm)	–	–	0.93***	(0.85, 0.99)
Period of time (exposure)	1			1
Model summary
LR $x^{2}$	600.34		542.12
$P$ > $x^{2}$	0.000		0.000
McFadden’s pseudo- $R^{2}$	0.212		0.213
Observations	319		319

Note:

Models 1 and 2 test the relationship between the number of accumulative cases $N$ as of 3 February 2020, the total number of new cases $Δ N$ during 4–11 February 2020, and variables, respectively. IMI from Wuhan refers to the IMI flows from Wuhan in the Two Weeks Before CNY (11–24 January 2020). ITI in Model 1 refers to the average ITI value during 18–27 January 2020. ITI in Model 2 refers to the average ITI value during 28 January to 4 February 2020. The analysis was done in Stata 16.0.

***P < 0.001, **P < 0.01, *P < 0.05.

CI: confidence interval; IMI: inter-city mobility index; ITI: intra-city travel intensity.

Urbanization and spatial heterogeneity

The regression coefficients in Model 1 show that in addition to IMI from Wuhan, the urbanization level also has significant impacts on epidemic severity. For every 10% increase in inter-city mobility volume and every one unit (1%) increase in the urbanization level, the number of infected cases in the city is expected to increase by 6.9% (95%CI: 6.4%–7.4%) and 1% (95%CI: 0.5%–1.5%), respectively. The urbanization level measures the percentage of people living in the urban area. In Chinese administrative units, urban area may account for only 7.7% of the administrative area in 2015 (Ma and Long, 2019). A higher urbanization level implies more people are living in the populated urban centers and may lead to a higher possibility of contacts between individuals. By contrast, population density in this study, which measures the population distribution in the whole administrative area (including urban and rural areas), has no significant influence on epidemic severity. In previous studies, the association between population density and disease transmission was found at the neighborhood-level (Grantz et al., 2016) but absent at larger spatial scales (Chowell et al., 2008; Viboud et al., 2006). It indicates that an average attribute of a large region may ignore the spatial heterogeneity between sub-regions and lead to an inaccurate result. Our results suggest that there may exist substantial heterogeneity in disease spread patterns between urban and rural areas.

Results in Model 1 show that IMI from Wuhan and urbanization level can well explain the incidence of the disease in a city. Values of McFadden’s pseudo-R² between 0.2 and 0.4 indicate extremely good model fits. This range is equivalent to 0.7 to 0.9 for a linear function in simulations (Louviere et al., 2000). Other variables, including intra-city mobility, are not important determinants of the disparities of disease incidences among cities. The extensive containment measures make the disease spread less subject to intra-city mobilities.

Humidity and mediated effects of other factors under containment interventions

We observed the negative impacts of absolute humidity on the disease spread in a city. The significance appears in Model 2 (local diffusion process) but is absent in Model 1 (transfer diffusion process). The regression coefficient shows that if the absolute humidity increases by one unit (1 g/km³), the number of new cases in a city will decrease by 4% (95%CI: 0.7%–7%). It suggests that warm and humid weather could help mitigate the spread of COVID-19. The conclusion is consistent with existing findings that cold and dry weather could drive the spread of some infectious diseases like inﬂuenza (Dalziel et al., 2018; Shaman and Kohn, 2009). Moreover, the coefficient of active cases suggests that if the active cases at a point of time increase by 10%, the new cases in the following week would increase by 9.3% (95%CI: 8.5%–9.9%) after holding other variables constant. Except for active cases and absolute humidity, other variables play negligible roles in explaining the variations of new cases in Model 2.

Conclusion and discussion

This study analyzed the interplay of spatial spread of COVID-19 and human mobility in the urban system of China during 2020 CNY. Furthermore, we compared the contribution of human mobility and other factors in the disease spread. We found that the disease spread has experienced two stages in Chinese cities: a major transfer (inter-city) diffusion before CNY and a subsequent local (intra-city) diffusion after CNY. Variations in disease incidence among cities are mainly driven by inter-city mobility flows from the epidemic center of Wuhan. Cities that are closer to the epidemic center and with more people living in the urban area will face higher risks of disease incidence (see Secondary Sources in the Disease Spread and Figure S3 in the Supplemental Material for a more detailed discussion). For a 10% increase in travel volume and a one unit (1%) increase in the percentage of people living in the urban area, the number of infected cases in the city is expected to increase by about 7% and 1%, respectively. Our results also suggest that every one unit increase (1 g/km³) in absolute humidity is linked with a 4% decrease in the number of new cases in a city. However, the change of humidity alone may not necessarily cause the decline of cases unless extensive containment interventions are employed. The relationship between the degree of intra-city mobility and the variation of epidemic severity is a complicated interaction. The more severe the outbreak, the more drastic intra-city travel restriction measures the government will adopt, and the more cautious people’s behavior will be, which may lead to lower intra-city mobility. The impact of epidemic severity in reducing intra-city mobility can be reflected in Chinese cities in the Two Weeks After CNY. Although reducing intra-city mobility after implementing intra-city travel restriction did not immediately stop the growth of the absolute number of new cases, it effectively slowed down the spread of the disease.

China has taken unprecedented measures in controlling inter-city and intra-city mobility to contain the outbreak. Almost all provinces (except Tibet) in mainland China have initiated the highest-level of emergency response by 25 January 2020. The suspension and reduction of the inter-city public traffic and implementation of travel restrictions have substantially reduced inter-city mobility. The extension of the holiday, stay-at-home order, closure of schools and non-essential services, and encouragement of working from home have effectively reduced intra-city mobility. Smart technologies and big data, such as Health QR Code that was initially launched on 11 February in Hangzhou City and later being adopted in most cities of mainland China, are used to detect people’s mobility trajectories. Public transportation facilities use the real-name registration system. Once a patient is diagnosed, people who have been contacted with him or her can be found and informed with the help of big data very quickly. The neighborhoods in the city also adopt different degrees of lockdown according to the degree of disease spread. These measures avoided many contacts between infected and susceptible population. On top of this, wearing masks, social distancing, and washing hands could help people to reduce the human to human transmission. The extensive tests, timely isolation and treatment could help to further identify the contagious people, cut off the chain of disease spread, and prevent and control the outbreak.

The extensive travel restrictions in China have reduced approximately 70% and 40% inter-city and intra-city mobility, respectively. These measures have helped China to control the spread of COVID-19 and gradually re-open the cities after 9 February 2020. The travel restriction was removed from Wuhan, the epidemic center, after 8 April 2020. Adopting place-specific inter- and intra-city travel control policies is an effective way to contain an outbreak before the vaccine is available. It has to be carried out together with other disease monitoring, control and preventive measures such as building early warning systems, providing timely and transparent information, taking social distancing measures, trajectory tracing and contact tracing, wearing masks, and washing hands. The findings of our study could provide valuable insights into understanding the dynamics of disease spread in the urban system and help to respond to another new wave of pandemic in China and other parts of the world (see Supplemental Material for the discussion of limitation).

Supplemental Material

sj-pdf-1-epb-10.1177_2399808320954211 - Supplemental material for The interplay of spatial spread of COVID-19 and human mobility in the urban system of China during the Chinese New Year

Supplemental material, sj-pdf-1-epb-10.1177_2399808320954211 for The interplay of spatial spread of COVID-19 and human mobility in the urban system of China during the Chinese New Year by Xiaoyan Mu, Anthony Gar-On Yeh and Xiaohu Zhang in EPB: Urban Analytics and City Science

Footnotes

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by the Chan To-Haan Endowed Professorship Fund of the University of Hong Kong.

ORCID iDs

Xiaoyan Mu

Xiaohu Zhang

Supplemental material

Supplemental material for this article is available online.

Xiaoyan Mu is a post-doctoral fellow in the Department of Urban Planning and Design at the University of Hong Kong. She received her Ph.D in Urban Planning from the University of Hong Kong. Her research interests include population migration, urban system development, and mega-region development in China.

Anthony Gar-On Yeh is chair professor in Urban Planning and Geographic Information Systems in the Department of Urban Planning and Design at the University of Hong Kong. His research interests include GIS, urban modelling, big data, smart cities, and urban planning and development in Hong Kong, China, and Southeast Asia.

Xiaohu Zhang is a post-poctoral Fellow at the MIT Senseable City Lab. He received his Ph.D in GIS and remote sensing from the University of Hong Kong. His scholarship bridges the information gap in sustainable urban and transportation policy-making with stochastic simulation and big data analytics. Broadly interested in urban data science, his recent work explores the sustainability of new shared mobility services, such as scooter sharing, carsharing and ridesharing.

References

Abdullah

, Thomas

, Mcghee

, et al. (2004) Impact of severe acute respiratory syndrome (SARS) on travel and population mobility: Implications for travel medicine practitioners. Journal of Travel Medicine 11(2): 107–111.

Bajardi

Poletto

Ramasco

, et al. (2011) Human mobility networks, travel restrictions, and the global spread of 2009 H1N1 pandemic. PLoS ONE 6(1): e16591.

Balcan

Colizza

Gonçalves

, et al. (2009) Multiscale mobility networks and the spatial spreading of infectious diseases. Proceedings of the National Academy of Sciences of the United States of America 106(51): 21484–21489.

Bedford

Cobey

Beerli

, et al. (2010) Global migration dynamics underlie evolution and persistence of human influenza A (H3N2). PLoS Pathogens 6(5): e1000918.

Belik

Geisel

Brockmann

(2011) Natural human mobility patterns and spatial spread of infectious diseases. Physical Review X 1(1): 011001.

Bengtsson

Gaudart

, et al. (2015) Using mobile phone data to predict the spatial spread of cholera. Scientific Reports 5: 8923.

Buckee

Balsari

Chan

, et al. (2020) Aggregated mobility data could help fight COVID-19. Science 368(6487): 145–146.

China CDC (2020) The epidemiological characteristics of an outbreak of 2019 novel coronavirus diseases (COVID-19). China CDC Weekly 2(8): 113–122.

China Internet Network Information Center (2019) The 44th China Statistical Report on Internet Development. Available at: http://www.cac.gov.cn/2019-08/30/c_1124938750.htm (accessed 10 March 2020).

10.

Chinazzi

Davis

Ajelli

, et al. (2020) The effect of travel restrictions on the spread of the 2019 novel coronavirus (COVID-19) outbreak. Science 368(6489): 395–400.

11.

Chowell

Bettencourt

Johnson

, et al. (2008) The 1918–1919 influenza pandemic in England and Wales: Spatial patterns in transmissibility and mortality impact. Proceedings of the Royal Society B: Biological Sciences 275(1634): 501–509.

12.

Colizza

Barrat

Barthélemy

, et al. (2006) The role of the airline transportation network in the prediction and predictability of global epidemics. Proceedings of the National Academy of Sciences of the United States of America 103: 2015–2020.

13.

Colizza

Barrat

Barthelemy

, et al. (2007) Modeling the worldwide spread of pandemic influenza: Baseline case and containment interventions. PLoS Medicine 4(1): 0095–0110.

14.

Dalziel

Kissler

Gog

, et al. (2018) Urbanization and humidity shape the intensity of influenza epidemics in U.S. cities. Science 362(6410): 75–79.

15.

Dalziel

Pourbohloul

Ellner

(2013) Human mobility patterns predict divergent epidemic dynamics among cities. Proceedings of the Royal Society B: Biological Sciences 280(1766): 20130763.

16.

Eubank

Guclu

Anil Kumar

, et al. (2004) Modelling disease outbreaks in realistic urban social networks. Nature 429(6988): 180–184.

17.

Findlater

Bogoch

(2018) Human mobility and the global spread of infectious diseases: A focus on air travel. Trends in Parasitology 34(9): 772–783.

18.

Gardner

Sarkar

(2013) A global airport-based risk model for the spread of dengue infection via the air transport network. PLoS ONE 8(8): e72129.

19.

Gardner

Bóta

Gangavarapu

, et al. (2018) Inferring the risk factors behind the geographical spread and transmission of Zika in the Americas. PLOS Neglected Tropical Diseases 12(1): e0006194.

20.

Grantz

Rane

Salje

, et al. (2016) Disparities in influenza mortality and transmission related to sociodemographic factors within Chicago in the pandemic of 1918. Proceedings of the National Academy of Sciences of the United States of America 113(48): 13839–13844.

21.

Grenfell and Bolker (1998) Cities and villages: Infection hierarchies in a measles metapopulation. Ecology Letters 1(1): 63–70.

22.

Guan

W-J

Z-y

, et al. (2020) Clinical characteristics of coronavirus disease 2019 in China. The New England Journal of Medicine 382(18): 1708–1720.

23.

Huang

Wang

, et al. (2020) Clinical features of patients infected with 2019 novel coronavirus in Wuhan, China. The Lancet 395: 497–506.

24.

Hufnagel

Brockmann

Geisel

(2004) Forecast and control of epidemics in a globalized world. Proceedings of the National Academy of Sciences of the United States of America 101(42): 15124–15129.

25.

Jia

Yuan

, et al. (2020) Population flow drives spatio-temporal distribution of COVID-19 in China. Nature 582: 389–394.

26.

Kraemer

MUG

Yang

C-H

Gutierrez

, et al. (2020) The effect of human mobility and control measures on the COVID-19 epidemic in China. Science 368(6490): 493–497.

27.

Lai

Ruktanonchai

Zhou

, et al. (2020) Effect of non-pharmaceutical interventions to contain COVID-19 in China. Nature. Epub ahead of print 4 May 2020. DOI: 10.1038/s41586-020-2293-x.

28.

Louviere

Hensher

Swait

(2000) Stated Choice Methods: Analysis and Application. Cambridge: Cambridge University Press.

29.

(2005) Urban administrative restructuring, changing scale relations and local economic development in China. Political Geography 24(4): 477–497.

30.

Long

(2019) Identifying spatial cities in China at the community scale. Journal of Urban and Regional Planning 11(1): 37–50 [in Chinese].

31.

Massaro

Kondor

Ratti

(2019) Assessing the interplay between human mobility and mosquito borne diseases in urban environments. Scientific Reports 9(1): 16911.

32.

Meloni

Perra

Arenas

, et al. (2011) Modeling human mobility responses to the large-scale spreading of infectious diseases. Scientific Reports 1(1): 62.

33.

Meng

Wang

Liu

, et al. (2005) Understanding the spatial diffusion process of severe acute respiratory syndrome in Beijing. Public Health 119(12): 1080–1087.

34.

Merler

Ajelli

(2010) The role of population heterogeneity and human mobility in the spread of pandemic influenza. Proceedings of the Royal Society B: Biological Sciences 277(1681): 557–565.

35.

National Bureau of Statistics of China (NBSC) (2015) Micro-data samples of the 2015 One percent population sample survey. Available at http://microdata.stats.gov.cn/ (accessed 10 March 2020).

36.

National Bureau of Statistics of China (NBSC) (2019) China City Statistical Yearbook 2018. Beijing: China Statistical Press.

37.

Quinn

(1994) Population migration and the spread of types 1 and 2 human immunodeficiency viruses. Proceedings of the National Academy of Sciences of the United States of America 91(7): 2407–2414.

38.

Russell

Jones

Barr

, et al. (2008) The global circulation of seasonal influenza A (H3N2) viruses. Science 320(5874): 340–346.

39.

Shaman

Kohn

(2009) Absolute humidity modulates influenza survival, transmission, and seasonality. Proceedings of the National Academy of Sciences of the United States of America 106(9): 3243–3248.

40.

Smallman‐Raynor

Cliff

(2001) Epidemic diffusion processes in a system of US military camps: Transfer diffusion and the spread of typhoid fever in the Spanish‐American War, 1898. Annals of the Association of American Geographers 91(1): 71–91.

41.

Sparks

Keighley

Muscatello

(2010) Early warning CUSUM plans for surveillance of negative binomial daily disease counts. Journal of Applied Statistics 37(11): 1911–1929.

42.

Tian

Cazelles

, et al. (2018) Urbanization prolongs hantavirus epidemics in cities. Proceedings of the National Academy of Sciences of the United States of America 115(18): 4707–4712.

43.

Tian

Liu

, et al. (2020) An investigation of transmission control measures during the first 50 days of the COVID-19 epidemic in China. Science 368(6491): 638–642.

44.

Viboud

Bjørnstad

Smith

, et al. (2006) Synchrony, waves, and spatial hierarchies in the spread of influenza. Science 312(5772): 447–451.

45.

World Health Organization (WHO) (2020) Report of the WHO-China Joint Mission on Coronavirus Disease 2019 (COVID-19). Available at: https://www.who.int/docs/default-source/coronaviruse/who-china-joint-mission-on-covid-19-final-report.pdf (accessed 10 May 2020).

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.52 MB