key: cord-0934789-2tcnvjf2
authors: Tian, T.; Luo, W.; Jiang, Y.; Chen, M.; Pan, W.; Zhao, J.; Yang, S.; Zhang, H.; Wang, X.
title: The timing and effectiveness of implementing mild interventions of COVID-19 in large industrial cities
date: 2020-06-23
journal: nan
DOI: 10.1101/2020.06.22.20137380
sha: 975be8b01d7064a508643986ee7c5e5153cbf18c
doc_id: 934789
cord_uid: 2tcnvjf2

The outbreak of novel coronavirus disease (COVID-19) has spread around the world since it was detected in December 2019. The Chinese government executed a series of interventions to curb the pandemic. The "battle" against COVID-19 in Shenzhen, China is valuable because populated industrial cities are the epic centres of COVID-19 in many regions. We made use of synthetic control methods to create a reference population matching specific characteristics of Shenzhen. With both the synthetic and observed data, we introduced an epidemic compartmental model to compare the spread of COVID-19 between Shenzhen and its counterpart regions in the United States that didn't implement interventions for policy evaluation. Once the effects of policy interventions adopted in Shenzhen were estimated, the delay effects of those interventions were referred to provide the further control degree of interventions. Thus, the hypothetical epidemic situations in Shenzhen were inferred by using time-varying reproduction numbers in the proposed SIHR (Susceptible, Infectious, Hospitalized, Removed) model and considering if the interventions were delayed by 0 day to 5 days. The expected cumulative confirmed cases would be 1546, which is 5.75 times of the observed cumulative confirmed cases of 269 in Shenzhen on February 3, 2020, based on the data from the counterpart counties (mainly from Broward, New York, Santa Clara, Pinellas, and Westchester) in the United States. If the interventions were delayed by 5 days from the day when the interventions started, the expected cumulative confirmed cases of COVID-19 in Shenzhen on February 3, 2020 would be 676 with 95% credible interval (303,1959). Early implementation of mild interventions can subdue the epidemic of COVID-19. The later the interventions were implemented, the more severe the epidemic was in the hard-hit areas. Mild interventions are less damaging to the society but can be effective when implemented early.

The confirmed cases of novel coronavirus (SARS-Cov-2) disease (COVID-19) had been detected since December 2019 1 , there are over 200 countries with reported confirmed cases 2 . A potent characteristic of SARS-Cov-2 is that it can be contagious during the incubation period, where asymptomatic individuals could become a . CC-BY-NC-ND 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 June 23, 2020. . https://doi.org/10.1101/2020.06.22.20137380 doi: medRxiv preprint 3 disseminating source 3 . Therefore, finding and implementing effective interventions is vital to disease control [4] [5] .

On January 23, 2020, nonpharmaceutical interventions including all public transportation closures started at Wuhan, and a similar response was triggered cities geographically near Wuhan cities, such as the cities within Hubei province and Wenzhou in Zhejiang province on February 1, 2020 6 . Guangdong province, where there is a large population of migrants from Hubei province 7 , activated the first-level public health emergency response on January 23, 2020 8 . Shenzhen, with a confirmed case of COVID-19 firstly reported in Guangdong province, implemented the interventions by closely following and isolating the close contacts of confirmed cases of COVID-19. By far, the strategic polices for direct protection (such as wearing face masks) and for transmission reduction (for example 14 days isolation for travelers, cancellation of public gathering, delayed reopening of schools) have been maintained in Shenzhen. The intervention and control strategies of Shenzhen were adopted as a classical example in the report of WHO 9 .

As an industrial city, where there is a large population of inbound and outbound travelers 10 , Shenzhen implemented a relatively mild intervention strategy compared with that of Wuhan and mounted an early response to the epidemic of COVID-19 relatively to other cities such as Wenzhou. Therefore, of great interest is to examine how effective such mild but early interventions were in curbing the epidemic of COVID-19 in Shenzhen. We evaluated the treatment effects of intervention policies in . CC-BY-NC-ND 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)

The copyright holder for this preprint this version posted June 23, 2020. 

Epidemic data were collected for daily cumulative confirmed cases of COVID-19 since January 19, 2020 in Shenzhen when the first confirmed case was reported 11 . Its corresponding population, GDP, and areas were collected in the statistical yearbook.

Also, the daily cumulative confirmed cases of COVID-19 for each county of each state in the United States were downloaded and available online 12 since March 1, 2020. The corresponding populations, areas, and latitudes were collected from the United States Census Bureau (USCB) 13-14 . The White House issued a "call to action" for coronavirus guidelines including canceling gathering over 10 people and staying at home on March 16, 2020 15 . Therefore, there were 16 days of duration before the interventions implemented in the United States. We compare the data between Shenzhen from January 19 to February 3, 2020 and counties in the United States from March 1 to March 16, 2020, by fixing the total duration at 16 days starting from the day of the first reported case in the locations of interest.

We employed synthetic control methods (SCM) to examine whether there were significant treatment effects of interventions implemented in Shenzhen from . CC-BY-NC-ND 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)

The copyright holder for this preprint this version posted June 23, 2020. . https://doi.org/10.1101/2020.06.22.20137380 doi: medRxiv preprint January 19, 2020, when the first confirmed case was reported, by comparing the epidemic situations in Shenzhen with the counties in the United States from March 1, 2020 when the first confirmed cases were reported in the most counties. The comparison was made for the duration of 16 days, because that was the duration for the period when the United States did not call for the stay-at-home policy from the first day when most counties reported their first cases. We consider hypothetically delaying the starting date of interventions from 0 to 5 days after January 23, 2020 when the interventions were implemented in Shenzhen and simulated the likely outcomes of delayed interventions by using time-varying reproduction numbers.

To apply synthetic control methods (SCM) 16 , we selected a pool of counties in the United States as a control region to Shenzhen (a treated region). In the control region, there were no interventions implemented for the COVID-19. The potential outcome of interventions in Shenzhen was expressed as: There were no broad interventions implemented in the early stage of the epidemic in the United States, which allowed us to define a pre-intervention period from March 1 to March 16. These 16-day epidemic data in the United States were . CC-BY-NC-ND 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 June 23, 2020. . https://doi.org/10.1101/2020.06.22.20137380 doi: medRxiv preprint 6 used as the benchmark in our comparison with the 16-day epidemic data in Shenzhen from January 19 to February 3, 2020. Thus, the values of t start from 1 to 16. The treatment effects of the interventions on the cases per 100,000 for Shenzhen were defined as 1

for Shenzhen in the post-interventions period. Reliable estimates of 0 1t Y could be obtained by constructing a synthetic region whose epidemic situation was similar to that in Shenzhen in the pre-intervention period 17 .

To select a pool of matching counties to Shenzhen, the pre-intervention features needed to be considered. We used the hierarchical clustering 18 to assign the counties of the United States into a group similar to Shenzhen based on population density and latitude. The selected counties were combined to a synthetic region by also considering the 2-day COVID-19 cases per 100,000 in the pre-intervention period.

The number of selected counties was determined by the epidemic situation and the sum of their corresponding weights was equal to 1. A placebo test also used to determine whether there are significant treatment effects of the interventions in Shenzhen 17 .

Besides the determination of treatment effects of intervention in Shenzhen, we considered the likely outcomes of the late implementation of interventions on different days. A common formulation of infectious disease is susceptible-exposed-infectious-removed (SEIR) model 19 . We refined the SEIR by . CC-BY-NC-ND 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 June 23, 2020. By the setting above, t R was calculated as:

where 1/ β is the mean of the incubation period of COVID-19, and is assumed to be . CC-BY-NC-ND 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 June 23, 2020. . https://doi.org/10.1101/2020.06.22.20137380 doi: medRxiv preprint

69 areas (68 counties in the United States) including Shenzhen were grouped by using their corresponding latitude and population density (supplementary appendix Table 1 ).

SCM was used to combine the 68 U.S. counties to construct a synthetic region of Shenzhen based on their latitude, population density and 2-day cases per 100,000 in the pre-intervention period. The synthetic and average values of latitude, population density, the third-day cases per 100,000 (i.e. January 21, 2020), and the fourth-day cases per 100,000 (i.e. January 22, 2020) were shown in Table 1 Table 2 .

Before the mild intervention policies were implemented on January 23, 2020, the trends of actual cases per 100,000 in both Shenzhen and "synthetic Shenzhen" were highly similar, suggesting that such synthetic region could be used to estimate the "counterfactual" results of Shenzhen. Based on the Figure 1 , after 1 day of the implementation of the policies, the growth rate of actual cases per 100,000 in Shenzhen was relatively slower than that of "synthetic Shenzhen" until January 29, 2020. After that, there was a sharp increase in the gap of cases per 100,000 between Shenzhen and "synthetic Shenzhen". On the 16 th day from January 19, 2020, i.e., . CC-BY-NC-ND 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 June 23, 2020. . https://doi.org/10.1101/2020.06.22.20137380 doi: medRxiv preprint February 3, 2020, the estimated number of cases per 100,000 in "Synthetic Shenzhen" was 11.72, which is 4.86 times of the actual Shenzhen (2.41) (Figure 1 ). In other words, Shenzhen had not implemented mild intervention policies on January 23, 2020, the projected number of cumulative confirmed cases would be reach 1307 on For those 68 counties, the probability of having a gap for Shenzhen under a random permutation of the control measures was 5%, which is statistically significant in a conventional test 16 . This suggested that the mild intervention policies of Shenzhen might have significantly reduced the COVID-19 cases per 100,000 ( Figure 2 ).

There were mild but early interventions implemented in Shenzhen. It is worth examining the likely outcomes of those interventions if they were delayed by different days (Figure 3 ).Based on our simulation by delaying the interventions, the expected . CC-BY-NC-ND 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 June 23, 2020. for different days of delay were summarized from January 29 to February 3, 2020 in Based on Fig 1, if the mild interventions delayed for 4 or 5 days, the epidemic of COVID-19 in Shenzhen could be more severe than that in synthetic Shenzhen (1307 cumulative confirmed cases on February 3, 2020).

We used the daily reported cumulative confirmed case data for 16 days since January 19, 2020, in Shenzhen and the corresponding data for 16 days since March 1, 2020 in 68 counties in the United States serving as a control group, i.e., the "synthetic Shenzhen," those 68 U.S. counties were selected to match Shenzhen by latitude, population density and the 2-day COVID-19 cases per 100,000 in the pre-intervention period. Those 68 U.S. counties did not implement systematic interventions.

The cases of COVID-19 per 100,000 in Shenzhen were significantly lower than . CC-BY-NC-ND 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 June 23, 2020. What is already known on this subject

• The interventions were used to control the fast spread of virus. There is a lack of studies from the timing of the interventions and the intensity of interventions for infectious disease . CC-BY-NC-ND 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 June 23, 2020. What this study adds

• Common and mild interventions in a representative Chinese city were compared with its counterpart areas in the United States where similar interventions were not implemented timely.

• The treatment effects of the interventions were evaluated by causal inference methods.

• The starting time of the interventions was important. The slower the interventions were implemented, the more severe epidemic of COVID-19

would have been in the hard hit areas.

. CC-BY-NC-ND 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 June 23, 2020. . CC-BY-NC-ND 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)

The copyright holder for this preprint this version posted June 23, 2020. . https://doi.org/10.1101/2020.06.22.20137380 doi: medRxiv preprint

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

National Health Commission of the People's Republic of China. Distribution of COVID-19 cases in the world 2020

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

The foot-and-mouth epidemic in Great Britain: pattern of spread and impact of interventions

The United States Census Bureau. Table of United States counties

The White House. 30 Days to slow the spread

Synthetic Control Methods for Comparative Case Studies: Estimating the Effect of California's Tobacco Control Program

Synth: An R Package for Synthetic Control Methods in Comparative Case Studies

Hierarchical grouping to optimize an objective function

Appropriate models for the management of infectious diseases

P-SIHR probabilistic graphical model: An applicable model of COVID-19 in estimating the number of infectious individuals without isolation and time-varying reproduction number

Incubation period of 2019 novel coronavirus (2019-nCoV) infections among travellers from Wuhan, China

International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity

We would like to thank all individuals who are collecting epidemiological data of the COVID-19 outbreak around the world.. CC-BY-NC-ND 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 June 23, 2020. . CC-BY-NC-ND 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 June 23, 2020. .

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 June 23, 2020. . CC-BY-NC-ND 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 June 23, 2020. . CC-BY-NC-ND 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 June 23, 2020. . CC-BY-NC-ND 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)The copyright holder for this preprint this version posted June 23, 2020. . https://doi.org/10.1101/2020.06.22.20137380 doi: medRxiv preprint