key: cord-0952314-d5l9yzzc
authors: Haruka, K.; Takizawa, A.
title: Human Mobility and Infection from Covid-19 in the Osaka Metropolitan Area
date: 2022-05-16
journal: nan
DOI: 10.1101/2022.05.12.22274931
sha: 7dfa550265170cc006e259343efb4ca4ee65bddd
doc_id: 952314
cord_uid: d5l9yzzc

Controlling human mobility is thought to be an effective measure to prevent the spread of the COVID-19 pandemic. This study aims to clarify the human mobility types that impacted the number of COVID-19 cases during the medium-term COVID-19 pandemic in the Osaka metropolitan area. The method used in this study was analysis of the statistical relationship between human mobility changes and the total number of COVID-19 cases after two weeks. In conclusion, the results indicate that it is essential to control the human mobility of groceries/pharmacies to less than 0% and that of parks to more than -20%. The most significant finding for urban sustainability is that urban transit was not found to be a source of infection. Hence governments in cities around the world may be able to encourage communities to return to transit mobility, if they are able to follow the kind of hygiene processes conducted in Osaka.

The COVID-19 pandemic was reported to positively impact urban sustainability in the short term [1] . However, the 24 pandemic might not continue to impact urban sustainability positively even in the medium term. For example, several 25 lockdowns caused changes in mobility, especially in public transportation, because of concerns about COVID-19 26 infection [2] . However, the modal shift from public transportations to cars will adversely effect on reduction of carbon 27 dioxide emissions [3] . Controlling human mobility is thought to be an effective nonpharmaceutical intervention to 28 prevent the spread of the COVID-19 pandemic worldwide, for example, in the United States, the EU, and China [4, 5, 6] . 29

To control human mobility, states of emergency have been declared many times during the medium term COVID-19 30 pandemic. For example, the Japanese government declared states of emergency several times in Osaka Prefecture [7] . 31

During these emergency declaration periods, the Subcommittee on Novel Coronavirus Disease Control in Japan 32 requested that residents reduce their human mobility by 50% [8] . During the first emergency declaration, this requirement 33 reduced the home range in suburban cities of the Osaka metropolitan area by 50% due to residents changing their 34 transportation methods to walking and cycling [9,10,11]. During the same first emergency declaration, a decrease in 35 human mobility was also reported in Tokyo [12] . However, some studies have suggested that the effects of controlling 36 human mobility differed in the early term and medium term [13, 14, 15] . 37 This study investigates the following research question: Where should we control human mobility to reduce the 38 number of COVID-19 cases? The effectiveness of controlling human mobility has been reported through the 39 implementation of school and company closures [9,10,11]. In addition, in Japan, many prefectural governments requested 40 that dining and drinking establishments close by 8:00 p.m. and refrain from serving alcohol [7] . However, to prevent the 41 spread of the COVID-19 pandemic, it is necessary to consider restricting places that have not previously been considered. 42

In addition, to maintain socioeconomic activities, it is essential to consider ending restrictions in places where infections 43 are less likely to occur. Therefore, the type of human mobility contributes to how policymakers develop policies to 44 control the spread of infection for urban sustainability. 45

This study aims to clarify the human mobility types that impacted the number of COVID-19 cases during the 47 Figure 2 shows daily changes in human mobility in the Osaka, Kyoto, and Hyogo Prefectures for the medium-term 100 COVID-19 pandemic. Figure 2 shows the spline curve and the confidence interval. The smoothing parameter of the 101 spline curve λ was set to 0.001. Additionally, Figure 2 indicates the emergency declaration period. The results show 102 similar changes in the Osaka, Kyoto, and Hyogo Prefectures. 103 Figure 2 shows that human mobility varies according to the six types. After March 2020, all types of human 104 mobility except human mobility of residential are decreased. This finding suggests that more people stayed at home, even 105 without the stay-at-home order. When human mobility of residential are increased, other types of human mobility 106 decreased. Due to the emergency declaration, human mobility decreased in transit stations and retail/recreation. In 107 addition, the human mobility of workplaces declined sharply during holiday periods, such as summer vacations and the 108 new year holiday. The human mobility of groceries/pharmacies remained at approximately 0%, although it changed 109 slightly during the emergency declaration. The human mobility of parks increased during the first emergency declaration 110 but then began to decrease. 111 Figure 3 shows the daily change in the number of people infected with SARS-CoV-2 in the Osaka, Kyoto, and 113

Hyogo Prefectures. Figure 3 shows the spline curve and the confidence interval. The smoothing parameter of the spline 114 curve λ was set to 0.001. Additionally, Figure 3 indicates the emergency declaration period. 115 Figure 3 shows that the Osaka, Kyoto, and Hyogo Prefectures experienced five waves of increases and decreases in 116 COVID-19 cases between February 2020 and December 2021. The first wave was from April to May 2020, the second 117 from July to September 2020, the third from December 2020 to February 2021, the fourth from March to June 2021, and 118 the fifth from July to September 2021. The number of infections gradually increased from the first to the fourth wave. 119

States of emergency were declared during the first, third, fourth, and fifth waves. The declaration of a state of emergency 120 effectively reduced the number of COVID-19 cases. 121

Table 1 and Figure 4 show the human mobility types that impacted the total number of COVID-19 cases after two 123 weeks in the Osaka, Kyoto, and Hyogo Prefectures. The statistical analysis was the random forest method. Table 1 shows 124 the main effect and the total effect for each prefecture. Figure 4 shows the variable importance in the Osaka, Kyoto, and 125

Hyogo Prefectures. As shown in Table 1 , the R 2 scores of all models were over 0.7, indicating good accuracy. The results 126 are discussed separately by prefecture. 127

In Osaka Prefecture, the R 2 score was 0.777, which indicates that the model has good accuracy. The total effect was 128 higher for human mobility in groceries/pharmacies (total effect=0.437), parks (total effect=0.368), workplaces (total 129 effect=0.253), and residential areas (total effect=0.234). The total number of COVID-19 cases after two weeks gradually 130 decreased by decreasing the human mobility of groceries/pharmacies by approximately 5% to -5%. In addition, the 131 human mobility of parks increased from -20% to 20%, which reduced the number of COVID-19 cases. It was also found 132 that the total effect was lower for human mobility in transit stations (total effect=0.102). 133 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 16, 2022. ;

In Kyoto Prefecture, the R 2 score was 0.821, which indicates that the model has high accuracy. The total effect was 134 higher for human mobility in groceries/pharmacies (total effect=0.418), parks (total effect=0.363), and residential areas 135 (total effect=0.216). The total number of COVID-19 cases after two weeks gradually decreased by decreasing the human 136 mobility of groceries/pharmacies by approximately 10% to 5%. In addition, the human mobility of parks increased from -137 20% to 50%, which reduced the number of COVID-19 cases. It was also found that the total effect was lower for human 138 mobility in transit stations (total effect=0.148). 139

In Hyogo Prefecture, the R 2 score was 0.775, which indicates that the model has good accuracy. The total effect was 140 higher for human mobility in groceries/pharmacies (total effect=0.495), parks (total effect=0.303), and residential areas 141 (total effect=0.245). The total number of COVID-19 cases after two weeks gradually decreased by decreasing the human 142 mobility of groceries/pharmacies by approximately 10% to 0%. In addition, the human mobility of parks increased from -143 50% to 10%, which reduced the number of COVID-19 cases. It was also found that the total effect was lower for human 144 mobility in transit stations (total effect=0.146). 145

In conclusion, the results of this analysis indicate that it is essential to control the human mobility of 148 groceries/pharmacies to less than 0% and the mobility of parks to more than -20%. This finding is important because 149 human mobility control would reduce the number of people infected with SARS-CoV-2. To control the human mobility 150 of groceries/pharmacies, the government must actively encourage residents to shop online and diversify the time spent 151 using grocery stores and pharmacies. In fact, during the Delta variant outbreak in the Osaka metropolitan area, there were 152 many incidents of infection clusters in groceries and department stores. Previously, the Japanese government did not 153 restrict the shopping necessary to maintain daily life, even during the emergency declaration period. This study suggests 154 that controlling the human mobility of groceries/pharmacies can prevent the rapid increase in the total number of cases 155 after two weeks in the emergency declaration period. The results differ from those of a previous study [22] . This finding 156 is significant because this study clarified the necessity of reducing the human mobility of grocery stores and pharmacies. 157

The target value for reducing human mobility is under 0%. 158

The human mobility of parks was also found to impact the number of infections. As previous studies have found 159 [21, 22] , increasing the human mobility of parks contributes to a decrease in the number of infections. This finding 160 suggests that increasing the human mobility of parks decreases the number of infections. It means that parks could be 161 actively used in the emergency declaration period instead of controlling human mobility in groceries/pharmacies. 162

The most significant finding for urban sustainability is that urban transit was not found to be a source of infection. 163

The human mobility of transit stations has been used as a reference in policymaking [20] . Hence governments in cities 164 around the world may be able to encourage communities to return to transit mobility if they can follow the kind of This study also clarified that the government needs to consider the third most influential type of human mobility 172 according to the characteristics of each prefecture. For example, the Osaka Prefecture government could reduce the 173 number of infections by increasing the human mobility of workplaces to 0% by allowing people to go to work. The 174 results suggest that it would be better for the government not to prevent people from going to work but rather to prevent 175 them from shopping and other activities associated with work. 176

Currently, many people can be vaccinated against the disease in many countries. On the other hand, new variants of 178 COVID-19 arise continuously. Therefore, it is unlikely that the pandemic will end, as the number of infections regularly 179 increases and decreases. Human mobility control may continue to be the most effective method of a nonpharmaceutical The limitation of this study was that it was able to analyze only six types of human mobility available on Google 192 Community Mobility Reports. Therefore, we cannot deny the possibility that the control of human mobility proposed by 193 this study might cause an increase in another type of human mobility and a gradual increase in the number of infections. 194

For example, would it truly be effective to restrict mainly dining and drinking establishments? To address this limitation, 195 future research should research more diverse types of human mobility using GPS location history data. These GPS log 196 data can be obtained at regular intervals from mobile phones with users' consent. Using such data, we can clarify the 197 relationship with the number of infections in more detail. 198 Google Community Mobility Reports chart movement trends by geography across six categories: retail/recreation, 203 groceries/pharmacies, parks, transit stations, workplaces, and residential areas. Retail/recreation includes restaurants, 204 cafes, shopping centers, theme parks, museums, libraries, and movie theaters. Groceries/pharmacies includes grocery 205 stores, food warehouses, farmers markets, specialty food shops, drug stores, and pharmacies. Parks includes local parks, 206 national parks, public beaches, marinas, dog parks, plazas, and public gardens. Transit stations includes public transport 207 hubs, such as subway, bus, and train stations. The data show relative changes in visitors to the six types of places 208 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 16, 2022. . In addition, all methods used in this study followed the "Guidelines for the Use of 213 Device Location Data," which prohibit the use of GPS data for any purpose that involves identifying individual users to 214 protect the privacy of users' GPS location history [37] . 215

This study analyzed the daily number of newly confirmed cases of SARS-CoV-2 in the Osaka, Kyoto, and Hyogo 217

Prefectures in the Osaka metropolitan area. The data were obtained from public information on COVID-19 infections 218 provided by the Japanese Ministry of Health, Labour and Welfare [38] . The published data include the number of new 219 COVID-19 cases each day by prefecture. The data do not include any personally identifiable information. 220

The random forest method was used to analyze the relationship between the total number of COVID-19 cases after 222 two weeks and the human mobility data. The random forest analysis of this study is an unsupervised analysis. Random 223 forest predicts a response value by averaging the predicted response values across many decision trees [39] . Each tree is 224 grown from a bootstrap sample of the training data. A bootstrap sample is a random sample of observations drawn with 225 replacement. In addition, the predictors are sampled at each split in the decision tree [40] . Compared to machine learning, 226 such as neural networks, random forests obtain highly accurate models with high R 2 scores. For the statistical analysis, 227 this study used JMP PRO 16.0. 228

The predictor variables are daily human mobility data of retail/recreation, groceries/pharmacies, parks, transit 229 stations, workplaces, and residential areas. The response variable is the total number of COVID-19 cases after two weeks 230 (fourteen days). The two-week lag is because SARS-CoV-2 takes approximately two weeks from infection to disease 231 occurrence [41] . The effectiveness of Google mobility data was validated for 10-day forecasts of COVID-19 cases [42] . 232

The number of trees in the forest is ten thousand for the random forest. 233

Based on the results of the random forest, this analysis focused on the R 2 score, the main effect, the total effect, and 234 the variable importance of the prediction profiler. The variable importance is assessed by the dependent resampled inputs, 235 which are factor values constructed from observed combinations using a k-nearest neighbors approach. 

. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. 

The Impact of Policy Measures on Human Mobility, COVID-19 Cases, and Mortality in the US: A 254 Spatiotemporal Perspective

The effect of large-scale anti-contagion policies on the COVID-19 pandemic

The effect of social distance measures on COVID-19 epidemics in Europe: an interrupted 258 time series analysis

Secretariat (office for COVID-19 and other emerging infectious disease control)

Prime minister of Japan and His cabinet

Impact of COVID-19 pandemic on home range in a suburban city in the 264 Osaka metropolitan area

Development of a spatio-temporal analysis method to support the prevention of COVID-19 infection: 266 Space-time kernel density estimation using GPS location history data in Urban Informatics and Future Cities

Changes in walkable streets during the COVID-19 pandemic in a suburban city in the 269 Osaka metropolitan area

Non-compulsory measures sufficiently reduced human mobility in Tokyo during the COVID-19 271 epidemic

Mobility restrictions were associated with reductions in COVID-19 incidence early in the pandemic: 273 evidence from a real-time evaluation in 34 countries

Observing the silent world under COVID-19 with a comprehensive impact analysis based on 275 human mobility

277 16. Google. Google COVID-19 Community Mobility Reports

Early pandemic COVID-19 case growth rates increase with city 279 size

Mobility network models of COVID-19 explain inequities and inform reopening

Mobility Change and COVID-19 in Japan: Mobile Data Analysis of Locations of Infection

The 67th Meeting of the Osaka Prefecture Task Force on COVID-19. (Reference) Changes in the 285 number of people staying at stations

No place like home: Cross-national data analysis of the efficacy of 288 social distancing during the COVID-19 pandemic

Causal graph analysis of COVID-19 observational data in German districts 290 reveals effects of determining factors on reported case numbers

Data-driven approach to understand the mobility patterns of the portuguese population 292 during the covid-19 pandemic

Early social distancing policies in Europe, changes in mobility & COVID-19 case trajectories: 294 Insights from Spring 2020

Mitigation policies, community mobility, and COVID-19 case counts in Australia

Impacts of COVID-19 pandemic on urban park visitation: a global 298 analysis

Outline of the Act Partially Amending the Act on Special Measures against Novel 300 Influenza, etc., and Other Relevant Acts

Guidelines for the use of device location data

Visualizing the data: information on COVID-19 infections

Predictive and Specialized Modeling: Bootstrap Forest: Overview of the Bootstrap Forest Platform

The Elements of Statistical Learning: Data Mining, Inference, and 331 Prediction

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

Public mobility data enables COVID-19 forecasting and management at local and global scales