key: cord-1013945-ieejlnk0
authors: Ng, T.-C. V.; Cheng, H.-Y.; Chang, H.-H.; Liu, C.-C.; Yang, C.-C.; Jian, S.-W.; Liu, D.-P.; Cohen, T.; Lin, H.-H.
title: Effects of case- and population-based COVID-19 interventions in Taiwan
date: 2020-08-19
journal: nan
DOI: 10.1101/2020.08.17.20176255
sha: 85ed6bb4d1703657fb8c47960e324612083108d8
doc_id: 1013945
cord_uid: ieejlnk0

In the first wave of the COVID-19 pandemic, broad usage of non-pharmaceutical interventions played a crucial role in controlling epidemics. However, the substantial economic and societal costs of continuous use of border controls, travel restrictions, and physical distancing measures suggest that these measures may not be sustainable and that policymakers have to seek strategies to lift the restrictions. Taiwan was one of the few countries that demonstrated initial success in eliminating the COVID-19 outbreak without strict lockdown or school closure. To understand the key contributors to the successful control, we applied a stochastic branching model to empirical case data to evaluate and compare the effectiveness of more targeted case-based (including contact tracing and quarantine) and less targeted population-based interventions (including social distancing and face mask use) in Taiwan. We found that case-based interventions alone would not be sufficient to contain the epidemic, even in a setting where a highly efficient contact tracing program was in place. The voluntary population-based interventions have reduced the reproduction numbers by more than 60% and have likely played a critical role at the early stage of the outbreak. Our analysis of Taiwan's success highlights that coordinated efforts from both the government and the citizens are indispensable in the fight against COVID-19 pandemic.

While the coronavirus disease 2019 (COVID-19) outbreaks currently show no signs of deceleration in the Americas, Africa, and Southeast Asia 7 , some countries successfully contained the first wave of the outbreak with the use of strong interventions such as strict lockdowns and border closures [2] [3] [4] [5] 8 . Taiwan was initially considered a high-risk country for the COVID-19 outbreak, given its close relationship with China economically and geographically. Nevertheless, six months after the outbreak, Taiwan has one of the lowest per capita incidence and mortality rates of COVID-19 in the world, and no local cases were confirmed between early April and mid-August 9 . Notably, the containment of COVID-19 in Taiwan was achieved without strict lockdown or school closure. 10 To prevent the healthcare system from being overwhelmed, Taiwan implemented the "containment-asmitigation" or elimination strategy during the early phase 8, 11 . This approach included border control, casebased interventions targeting COVID-19 patients, and population-based measures targeting the general public 11, 12 . The case-based interventions included case detection and isolation through sensitive surveillance systems, contact tracing of confirmed cases to early detect secondary cases among close . 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 August 19, 2020. . https://doi.org/10.1101/2020.08.17.20176255 doi: medRxiv preprint contacts, and 14-day quarantine of close contacts (regardless of symptoms). The population-based measures included (mostly voluntary) face mask use, personal hygiene, and physical distancing 10, 13 . Despite Taiwan's initial success, it remains unclear which components among these various interventions contributed significantly to containment. Better understandings of the intervention effectiveness would not only help the preparedness for the next wave but also assist with phased control measures, especially during re-opening [1] [2] [3] 14 . This post hoc evaluation is particularly crucial as strict border control and drastic physical distancing has considerable short-term and long-term socio-economic repercussions. Several modeling studies examined the effectiveness of case-based or population-based interventions. However, most of them focused on the "what-if" analyses through simulating hypothetical scenarios without empirical linkage to specific real-world settings and primary data [14] [15] [16] [17] [18] . Using the epidemiological data and detailed contact tracing information in Taiwan, we applied a stochastic branching model with two-stage calibration to evaluate the effective reproduction number under case-based interventions (R_c), population-based interventions (R_p), and both (R_pc) and to quantify the impact of case-based interventions and population-based measures on the containment of the epidemic.

The epidemic started with few imported cases from China, leading to non-sustained local transmission during January-February 2020 (Fig 1a) . In March, a surge of imported cases mainly from North America and Europe caused sporadic local transmission. The capacity of laboratories testing for COVID-19 with reverse transcription-polymerase chain reaction (RT-PCR) was initially low and thus diagnostic testing was limited to symptomatic presumptive cases with relevant travel or contact history (Fig 1a) 11 The criteria of notification and testing were gradually expanded to include individuals with respiratory or alarming symptoms, with an overall test positive rate of 0.61% by June 1.

We analyzed the epidemiological and contact tracing data of 158 confirmed cases to estimate the incubation period, serial interval, and onset-to-isolation interval (Methods) 19 . The estimated mean incubation period and mean serial interval was 5.50 (95% credible interval [CrI]: 1.06-13.45) and 5.86 (95% CrI: -0.64 to 21.51) days, respectively (Fig. 1b) . The mean onset-to-isolation interval was 5.02 (95% CrI: -0.81 to 20.51) days, with a decreasing trend over time (Fig. 1c) . By fitting the stochastic . 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 August 19, 2020. . https://doi.org/10.1101/2020.08.17.20176255 doi: medRxiv preprint branching model to the observed serial intervals (Methods), we estimated that 52% (95% CrI: 39-67%) of transmission events occurred during the pre-symptomatic stage.

We used the fitted stochastic branching model and parameter values estimated from the empirical data to simulate the potential impact of case-based interventions (Table 1 for parameter values, Supplementary Notes for details of cased-based interventions). We found that the combination of case detection, contact tracing, and 14-day quarantine of close contacts (regardless of symptoms) could lower the reproduction number under case-based interventions (R_c) from the counterfactual value of 2.50 (R0) to 1.25 (95% CrI:

1.22-1.28) (Fig. 2a) . With 100 initial cases introduced to the community (i.e., cases escaping from border control and quarantine), the estimated probability of epidemic extinction was 0% (95% CrI: 0-0%). In the one-way sensitivity analysis, the value of onset-to-isolation interval had the most significant impact on the R_c, followed by the incubation period, the proportion of pre-symptomatic transmission, and the counterfactual R0 (Extended Data Fig. 1) . Notably, the extinction probability was consistently 0% when the counterfactual R0 was set to 2-3. When the onset-to-isolation interval was shortened to the lowest biweekly value of 2.5 days observed in Taiwan (Fig. 1c) , the R_c was 1.04 (95% CrI: 1.01-1.07), and the extinction probability was still 0% (95% CrI: 0-0%).

Among the different case-based interventions, quarantine of contacts contributed the most to the reduction of R_c (Fig. 2a) . The case-based interventions had only limited impacts on secondary transmissions, but tertiary and quaternary transmission events could still be reduced (Fig. 2b) , and resulted in a lower probability of asymptomatic transmission (Fig. 2c) .

. 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 August 19, 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 August 19, 2020. . https://doi.org/10.1101/2020.08.17.20176255 doi: medRxiv preprint . 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 August 19, 2020. . https://doi.org/10.1101/2020.08.17.20176255 doi: medRxiv preprint 

Since case-based interventions alone were not sufficient, population-based interventions must also have contributed to the containment. We firstly fitted the model to the observed cluster size distribution to infer the reproduction number under population-based interventions only (R_p), assuming a uniform prior . 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 August 19, 2020. . https://doi.org/10.1101/2020.08.17.20176255 doi: medRxiv preprint distribution (0.10-3.00) of R_p. The posterior estimate of R_p was 0.70 (95% CrI: 0.52-0.90), suggesting a 62%, 69%, and 75% reduction when the counterfactual R0 was 2.00, 2.50, and 3.00 respectively (Extended Data Fig. 2 ). The posterior estimate of R_pc using the fitted model was 0.49 (95% CrI: 0.41-0.59), which was similar with the R_pc directly estimated from the average size of clusters (R_pc: 0.36 [95% CrI: 0.14-0.50]). We note that this estimation procedure of R_p and R_pc was not influenced by the value of counterfactual R0. Table 1 ). The estimated Rt of influenza in 2020 dropped from 0.87 on January 21 (when the first case COVID-19 was reported) to 0.20 one month later, corresponding to a 77% decline. Analysis of estimated influenza incidence showed a similar pattern, with a 46% Rt reduction from 1.07 on January 21 to 0.58 after January 21 (Fig. 3b ).

We projected the epidemic curve with 100 initial cases under different scenarios using the fitted model (R0: 2.5, R_c: 1.25, R_p: 0.70, R_pc: 0.49). The case-based interventions could partially suppress the rate of increase, but exponential growth would continue and the daily number of new cases would rise to 5,617 (95% CrI: 3,265-8,335) by day 60 (Fig. 4a) . In contrast, we estimated that the effects of population-based interventions would be sufficiently large to control the epidemic and lead to local extinction, provided that behavioral changes could be maintained. Combining case-based and populationbased interventions would more quickly control the epidemic and provides insurance in case of lapses.

We further estimated the probability of successful COVID-19 containment in Taiwan with case-based interventions under different values of R_p and initial numbers of introductions (representing variation in the prevalence of infection in travelers and the effectiveness of border controls with corresponding quarantine requirement) (Fig. 4b) . When the number of introductions was set to greater than 100, R_p had to be kept below 1.6 to achieve an extinction probability of at least 90%. If R_p was greater 2.5, it would be impossible to rely only on case-based interventions to contain the outbreak (extinction probability: 0) even when the number of introductions was small.

. 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 August 19, 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 August 19, 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 August 19, 2020. 

Using a flexible modeling approach that incorporated multiple sources of primary data, we estimated and compared the impact of case-based and population-based interventions in Taiwan. We found that casebased interventions alone were not sufficient to contain the epidemic, even in a setting where the public health and healthcare system was not yet overwhelmed, and a highly efficient contact tracing program was in place. We also found that the mostly voluntary population-based interventions reduced the reproduction numbers by more than 60% and played a crucial role in the containment in Taiwan. Our analysis of Taiwan's success highlights that coordinated efforts from both the government and the citizens are indispensable in the fight against COVID-19 pandemic.

. 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 August 19, 2020. . https://doi.org/10.1101/2020.08.17.20176255 doi: medRxiv preprint Previous modeling studies showed promising effectiveness of contact tracing with corresponding management of contacts (either quarantine or active monitoring) in settings with well-functioning public health programs like Taiwan [15] [16] [17] . However, our results suggested that even in a well-prepared setting, contact tracing alone would fail to eliminate an epidemic when multiple introductions were likely. This discrepancy is driven primarily by the growing understanding of the role of pre-symptomatic transmission and the challenge this presents for shortening the delay from symptom onset to isolation. The effectiveness of contact tracing depends on the timeliness of case detection and implementing corresponding quarantine for high-risk contacts 16 . When the timeliness of case-based interventions has been optimized, the remaining transmissions can be blocked only by implementing more generalized nonsymptom-based approaches. According to our analysis of local data, the mean onset-to-isolation interval was around five days in Taiwan. This relatively long delay compared to the short serial interval suggests the virus usually had already been transmitted in the community by the time locally-acquired infections were detected. Therefore, population-based interventions should be prioritized over case-based interventions, even in settings where case-based interventions are feasible. That said, we did find casebased interventions to have significant impact on their own, and they provide additional assurance that the epidemic can be controlled and brought to extinction more quickly and reliably than under use of only population-based interventions.

On the other hands, we found that population-based interventions played a major role in containment. A meta-analysis of 172 observational studies in healthcare and non-healthcare settings revealed that physical distancing, face mask use, and eye protection were significantly associated with reduced COVID-19 transmission at the individual level 23 . The population-level effects of these interventions were assessed by other scenario-based analyses without empirical evaluation 3, [15] [16] [17] 24 Our study is among the very few that directly and empirically evaluated the population-level impact. In Taiwan This study provides crucial insights into the role of different control measures at different stages and settings. Border control with corresponding quarantine for entering travelers may be an option to limit the epidemic in the beginning by reducing the number of introductions, especially in certain contexts (e.g., .

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 August 19, 2020. . https://doi.org/10.1101/2020.08.17.20176255 doi: medRxiv preprint island nations like Taiwan and New Zealand) 8 . In Taiwan, the quarantine requirement for travelers returning from other countries after March 21 effectively limited the number of introductions in the community. However, the effect of border control with quarantine would be quickly diminished when the number of introductions exceeds 100 (Fig. 4b) , and this has important implications for the possibility of new epidemics in places like Taiwan that have achieved successful control over the first wave. This finding might also explain the resurgences that occurred in Hong Kong, Vietnam, and New Zealand when travel restrictions were lifted after a period of epidemic 25, 26 . With a sufficiently large number of introductions, the probability of local transmission is high, and containment efforts should shift toward suppressing local transmission using both case-based and population-based interventions. Our analysis revealed that blended approaches could reliably result in epidemic extinction, even when the number of introductions is high. When discussions on re-opening borders to boost economic growth raise, our results suggest that travel restrictions may be judiciously lifted only when other effective interventions can be maintained. The stringency of border control can be calibrated by pre-or post-arrival testing and entry quarantine, depending on the infrastructure capacity of individual countries and the number of COVID-19 cases that can be managed by local public health and healthcare system.

We note an interdependency between case-based and population-based interventions 15 . For example, contact tracing and quarantine would be increasingly difficult in practice when the number of cases surges. In this case, the intensity of population-based interventions has to be strengthened to compensate for the decreased efficiency of case-based interventions. On the other hand, if the population-based interventions can effectively suppress local transmission, it would require less effort from the public health workforce to contain the community outbreak. Nonetheless, maintaining behavioral changes and adopting the new normal is challenging in many places 27 . Since the health behaviors may change rapidly over time, monitoring these behaviors in the community would be informative 6 . In our analysis, we explored the potential of using influenza activity as a proxy indicator to monitor the Rt 28 . Continuous measurement of the implementation and impacts of case-based and population-based interventions would help to calibrate the efforts in contact tracing and quarantine and to reserve the precious public health resources.

One limitation of this study is that we could not directly estimate the "counterfactual" R0 in the absence of any interventions since the majority of case-based and population-based interventions were triggered soon after the outbreak begun. Nonetheless, our main conclusions were robust within the range of commonly reported R0 values of 2-3. Additionally, the analysis of cluster size distribution relied on the assumption of complete information of all clusters. If small clusters were more likely to be missed by surveillance and . 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 August 19, 2020. . https://doi.org/10.1101/2020.08.17.20176255 doi: medRxiv preprint contact tracing, the estimated reproduction number (R_pc) would have been lower, and the impact of population-based intervention would have been larger.

Through the analysis of a presumably high-risk country, we found that the combination of case-based interventions from an effective public health system and voluntary behavioral changes in a vigilant population likely contributed to the initial success of epidemic control. Notably, case-based interventions alone may not be sufficient to contain COVID-19, given the significant role of pre-symptomatic transmission. Population-based interventions were the key to success in our setting. Still, the longer-term sustainability and generalizability of these behaviors (e.g., face mask use) to other settings must be assessed.

Case series data of SARS-CoV-2 infections in Taiwan were collected from the official website of Taiwan Centers for Disease Control (TCDC) and reviewed by TCDC officers to clarify missing information. All the cases were confirmed by RT-PCR tests using one of the following clinical specimens: nasopharyngeal swab, throat swab, expectorated sputum, or lower respiratory tract aspirates 29 . Cases were isolated immediately after being notified to Taiwan CDC.

We analyzed the epidemiological and contact tracing data to capture the transmission dynamics of COVID-19 in Taiwan 19 . Starting March 21 all inbound passengers (citizens and eligible non-citizens) to Taiwan were required to undergo a 14-day quarantine upon entry, and nearly all confirmed cases afterward were imported and were diagnosed during the quarantine period (Supplementary Notes). We, therefore, enrolled locally-acquired cases, epidemiologically-confirmed clusters, and imported cases who entered Taiwan before March 21 and excluded the returnees who were tested at the airport or diagnosed during home quarantine (final sample size: 158 cases with 3154 contacts).

We adapted the stochastic branching process model developed by Hellewell 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 August 19, 2020. . https://doi.org/10.1101/2020.08.17.20176255 doi: medRxiv preprint the sampled generation interval and the isolation or quarantine period of the index case (Extended Data   Fig. 4) . We parameterized the distribution of generation interval as a skewed normal distribution centered at each index case's onset time to avoid discordant incubation period and generation interval. The shape of generation interval distribution was determined by the proportion of pre-symptomatic transmission and the standard deviation of generation interval. Distributions of the incubation period, serial interval, and onset-to-isolation interval were independently estimated from data using Bayesian hierarchical models (as described in Supplementary Method 1, Supplementary Table 1 for details of the branching model and parameterization). Fig. 4 Examples of the effects of case detection, contact tracing, and quarantine. Case A and A* demonstrate the effect of mere detection, which can only prevent the transmission once the active cases are tested and isolated. That is, the active cases can transmit the disease during their incubation and . 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 August 19, 2020. . https://doi.org/10.1101/2020.08.17.20176255 doi: medRxiv preprint delay of case detection. Case B and B* demonstrate the effect of detection plus tracing (without quarantine), where case B was successfully traced and onset within a buffer period. Therefore, case B was immediately isolated when the source was detected. Detection plus tracing can prevent transmission during the delay of case detection. Case C and C* demonstrate the combined effect of detection, contact tracing, and quarantine. Only in this scenario that transmission during the incubation period can be prevented. We assume that there is no delay between testing and isolation, the buffer of contact tracing to be one day, and the same effect of quarantine and isolation. Besides, asymptomatic cases are never detected or traced but could be quarantined and have lower transmissibility.

We estimated the unknown key parameters by fitting the branching model to empirical data using twostage calibration with the sequential Monte Carlo algorithm 15, 30 

We estimated the effect of case-based interventions using the stochastic branching model after stage-1 fitting. Since the majority of these containment measures were in place at the beginning of the epidemic, it was not possible to directly estimate the "counterfactual" R0 (the hypothetical reproduction number without interventions) in Taiwan. We assumed this counterfactual R0 to be 2.50 (range 2-3), similar to the estimated R0 in Hong Kong at the beginning of its outbreak and consistent with the previously estimated R0 values 6,31,32 . Five scenarios were considered: (1) no case-based interventions; (2) case detection and isolation; (3) case detection and contact tracing to detect and isolate secondary cases; (4) case detection, contact tracing, and 7-day quarantine for contacts (regardless of symptoms), (5) case detection, contact tracing, and 14-day quarantine (Supplementary Table 3 ). Whereas the 14-day quarantine for contacts was a current policy implemented in Taiwan, we simulated the scenario with 7-day quarantine to see the trend of impact of quarantine duration. The primary indicator was the mean effective reproduction number, along with the probability of outbreak extinction, which was defined as zero new cases within 20 generations.

. 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 August 19, 2020. . https://doi.org/10.1101/2020.08.17.20176255 doi: medRxiv preprint

The effect of population-based interventions was estimated using two independent approaches. First, we analyzed the observed cluster size distribution in Taiwan 

The datasets generated and analyzed during the current study are available from the corresponding author on reasonable request.

Codes are available on GitHub at https://github.com/dachuwu/Taiwan_CovidDTQ before publication.

Effect of non-pharmaceutical interventions to contain COVID-19 in China

Estimating the effects of non-pharmaceutical interventions on COVID-19 in Europe

Effects of non-pharmaceutical interventions on COVID-19 cases, deaths, and demand for hospital services in the UK: a modelling study

Reconstruction of the full transmission dynamics of COVID-19 in Wuhan

The effect of travel restrictions on the spread of the 2019 novel coronavirus (COVID-19) outbreak

Impact assessment of non-pharmaceutical interventions against coronavirus disease 2019 and influenza in Hong Kong: an observational study. The Lancet Public Health

Coronavirus disease (COVID-19) Situation Report -192. Coronavirus disease (COVID-19) Situation Report -192

Successful Elimination of Covid-19 Transmission in New Zealand

COVID-19 in Taiwan

Safeguarding the community line of defense to create a positive new lifestyle for disease prevention

Initial rapid and proactive response for the COVID-19 outbreak -Taiwan's experience

Policy Decisions and Use of Information Technology to Fight Coronavirus Disease

Response to COVID-19 in Taiwan

Effectiveness of isolation, testing, contact tracing, and physical distancing on reducing transmission of SARS-CoV-2 in different settings: a mathematical modelling study

Individual quarantine versus active monitoring of contacts for the mitigation of COVID-19: a modelling study

Impact of delays on effectiveness of contact tracing strategies for COVID-19: a modelling study

Feasibility of controlling COVID-19 outbreaks by isolation of cases and contacts

A modelling framework to assess the likely effectiveness of facemasks in combination with 'lock-down' in managing the COVID-19 pandemic

Contact Tracing Assessment of COVID-19 Transmission Dynamics in Taiwan and Risk at Different Exposure Periods Before and After Symptom Onset

Presymptomatic SARS-CoV-2 Infections and Transmission in a Skilled Nursing Facility

Evidence Supporting Transmission of Severe Acute Respiratory Syndrome Coronavirus 2 While Presymptomatic or Asymptomatic

Collateral Benefit of COVID-19 Control Measures on Influenza Activity

Physical distancing, face masks, and eye protection to prevent person-to-person transmission of SARS-CoV-2 and COVID-19: a systematic review and meta-analysis

Impact of non-pharmaceutical interventions (NPIs) to reduce COVID-19 mortality and healthcare demand

The mysterious resurgence of Covid-19

Four cases of COVID-19 with unknown source

The role of community-wide wearing of face mask for control of coronavirus disease 2019 (COVID-19) epidemic due to SARS-CoV-2

Using influenza surveillance networks to estimate state-specific prevalence of SARS-CoV-2 in the United States

Guidelines for COVID-19

Comparing nonpharmaceutical interventions for containing emerging epidemics

The reproductive number of COVID-19 is higher compared to SARS coronavirus

Projecting the transmission dynamics of SARS-CoV-2 through the postpandemic period

Measles Outbreaks in a Population with Declining Vaccine Uptake

Real-Time Surveillance of Infectious Diseases: Taiwan's Experience

Taiwan National Infectious Disease Statistics System

The authors declared no competing interests.