Optimal Social Distancing Policy for COVID-19 Control in Korea: A Model-Based Analysis

Since March 2020, when coronavirus disease 2019 (COVID-19) was declared a pandemic, many countries have implemented unprecedented restrictive measures to contain the spread of the virus.1 Due to the economic and social damage, governments cannot continue lockdowns indefinitely, and therefore, sustainable and public health-driven exit strategies are required. With limited evidence guiding governments to contain the virus optimally, many countries have taken various measures, including vaccinating, testing, tracing, quarantining, and social distancing,2 to slow the spread by ‘flattening the curve.’ South Korea is one of the few countries in the world to have successfully controlled the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) transmissions by implementing the “3T strategy” (tracing, testing, and treating) and social distancing with a mask mandate without a strict lockdown.3 South Korea collected an enormous amount of precise empirical data through active contact tracing, testing and vaccination, including the variants associated with infections, diagnoses and age-specific symptom onset dates, and detailed information (age, types, doses) on the administered COVID-19 vaccines.4 Since more than 70% of the total population received a second dose of vaccine,5 the government announced an exit strategy with a gradual return to normal life beginning November 2021.6

Meantime, the spread of highly contagious variants globally, including Delta and Omicron, has made it more difficult for health authorities to curb the spread, despite the increasing vaccinations and stringent social distancing over several months. While increasing vaccinations are expected to have an important contribution in returning life to normal, it may not be the only part of an exit strategy as numerous issues and uncertainties concerning the existing COVID-19 vaccines with emerging new variants.7 The quality and length of protection the vaccines will provide and their effectiveness in stopping viral transmission are yet to be determined. Supply-chain constraints, pricing, and unequal vaccine procurement across countries mean that coverage across most countries will remain below the level required for herd immunity.7 While each intervention is expected to reduce transmission, it is difficult to assess the relative or collective impact of multiple interventions under different conditions. Optimal strategies may be different by epidemic conditions (e.g., stage of epidemic curves, transmissibility of virus, population susceptibility), country-specific context (e.g., vaccination coverage, health system capacity), and policy choice (e.g., types and level of social distancing control).

Quantifying the impact of a particular or collective strategy under varying conditions is thus an essential first step to promote social consensus and policy decision-making among competing interests and choices. South Korea, in the settings of relatively low incidence and high vaccination coverage with high health systems capacity/access, one of the critical debates for COVID-19 exit strategies is to determine how long and to what extent social distancing control should be implemented to mitigate epidemic burden and social and economic damage. This study aims to explore the optimal social distancing policy for COVID-19 control in South Korea to reopen society safely. We considered comprehensive intervention scenarios such as testing intensity, vaccination rate, and social distancing measures under uncertain future transmissibility conditions after lifting social distancing measures based on rich empiric data.

Based on the calibrated model representing the COVID-19 epidemic trends in South Korea between 1 July and 20 November 2021, we projected the future epidemic curves after 21 November 2021 for the five scenarios (Table 1). Assuming that new variants increase transmissibility 3–4 times from the current level, our findings suggest that an epidemic resurgence may be expected with a lift of social distancing control depending on the level and duration of the social distancing control as well as the extent of testing intensity and effective vaccination uptake rate. First, three months of social distancing control (assuming a reduction in transmissibility by 50% between 21 November 2021 to 30 March 2022) with the current level of testing intensity and vaccination uptake rate may result in a gradual drop in the epidemic curve but a high epidemic resurgence after March 2022 (Scenario 1), while four months of strict social distancing control (assuming a reduction in transmissibility by 60%) may result in an immediate and sustained drop in the epidemic curve (Scenario 2), averting 1,327,000 cases and 1,766 ICU admissions compared to the base case. In contrast, despite increased testing intensity and vaccinations rates by 20%, two months of strict social distancing control may result in an epidemic resurgence, averting 705,000 cases and 1,191 ICU admission compared to the base case (Scenario 3). Four months of weak social distancing control (assuming a reduction in transmissibility by 40%) may result in an increased incidence between 21 November 2021 to 30 March 2022, if testing intensity and vaccination rate remain constant to the current level (Scenario 4), averting 363,000 cases and 748 ICU admissions compared to the base case. Even if the testing intensity and vaccination uptake rate are increased by 20%, two months of weak social distancing control may result in a high epidemic resurgence with a lift of social distancing control (Scenario 5), having additional 746,000 cases and 1,088 ICU admissions compared with the base case scenario.

The prevalence of severe patients was compared to the country’s ICU capacity (n = 1,500) for COVID-19 patients (Fig. 3). Increasing transmissibility of new variants and reduced social distancing level/duration may result in a substantial number of severe patients exceeding the current ICU capacity. We found four months of strict social distancing control or strong two months social distancing control with increased testing intensity and vaccination uptake rate may contain the ICU demand below the capacity limit (Scenario 2 and 3). On the other hand, moderate or weak social distancing control may result in substantial numbers of severe patients which are close to or exceed the current ICU capacity (Scenarios 1, 4, and 5) regardless of the duration of control (2–4 months) or testing intensity and vaccination increase by 20%.

The impact of all combination scenarios is presented in Fig. 4 as the cumulative incidence (Fig. 4A) and the peak number of severe patients (Fig. 4B). In general, the epidemic burden is greater as the levels of social distancing control, testing intensity and vaccination rate decrease, and transmissibility increases, as shown a change of cumulative incidence ranging from 393,000 to 4.4 million and the peak number of severe patients ranging from 394 to 10,000 between July 2021 and December 2022. Four months of strict social distancing control with constant testing intensity and vaccination uptake rate resulted in a cumulative incidence of 408,000 cases (under the condition of transmissibility increase three times of the current level) from 1 July 2021–30 December 2022. However, reducing to two months of strict social distancing control and increasing testing intensity and vaccination uptake rate by 20% may achieve a similar level of cumulative incidence (414,000 cases). However, with four months of strict social distancing control, ICU capacity is never to be breached unless both testing intensity and vaccination uptake rate decrease by 20% under increased transmissibility four times the current level. Meanwhile, two months of social distancing control will result in ICU capacity being more likely breached unless both testing intensity and vaccination uptake rate increased by 20% under strict social distancing enforcement.

This modelling analysis explored different intervention scenarios for an optimal COVID-19 exit strategy to reduce social distancing measures while minimising the epidemic burden, social and economic damage. After lifting social distancing control, our findings indicate a potential epidemic surge in three months (> 10,000 incidence cases). Four (or more) months of social distancing may mitigate a future epidemic resurgence and ICU demand keeping it below the capacity limit in both strong or weak social distancing levels if testing intensity and vaccination rate remain constant or increased by 20% of the current level. On the other hand, two months of strict social distancing may also successfully mitigate future epidemic surge and ICU demand as long as testing intensity and vaccination rates increase by 20% than the current level. Increasing testing intensity and vaccination uptake rate may help reduce necessary social distancing levels and duration to prevent future epidemic resurgence.

In South Korea, at the beginning stage of increasing incidence trend with emerging new variants, duration and level of social distancing greatly influence the future epidemic trend. Increasing the testing intensity and vaccination rates may help reduce the level and duration of mandatory social distancing to prevent the epidemic resurgence. Reducing the current testing intensity may result in an immediate small drop in the number of cases, but it would allow ongoing transmission by infected but undetected patients, and the number of cases will eventually increase over time, with an epidemic surge in June 2022. Similarly, reducing the current vaccination uptake rate may result in an epidemic surge in July 2022 due to the increase in the unvaccinated population. The impact and benefit of vaccination or testing are, however, greater, especially when the duration of social distancing is short (2 months) and the level of the control is strong (414,000 with 20% increased testing and vaccination vs. 2.6 million with 20% decreased testing and vaccination under three times increased transmissibility condition). In other words, in settings where and when testing intensity and vaccination uptake rate are relatively high, increasing social distancing duration from two to four months may result in marginal reduction of cumulative incidence (5% from 414,000 to 393,000 under strong control measure; 32% from 1.3 million to 880,000 with weak control measure). However, in settings wherein testing intensity and vaccination uptake rate are relatively lower, increasing social distancing duration from two to four months may reduce cumulative incidence substantially (79% from 2.5 million to 519,000) with strict social distancing enforcement.

Our findings are consistent with other modelling studies which have explored exit strategies, highlighting the critical considerations of a low reproduction number18 and timing/duration of lockdown.26 Dickens et al.26 using an agent-based model in Singapore, found that the lockdown start time has a greater impact than its duration and suggested that three months gradual release exit strategies are critical in suppressing the epidemic. Marziano et al.27 suggested controlling the reproduction number and incidence of infection are crucial. Social distancing/lockdown interventions have proven successful in curbing the spread of the disease.28, 29 Our study also showed the duration and level as the key drivers of the epidemic impact (Fig. 4). Nonetheless, the decision around the optimal duration/level of social distancing should be carefully considered based on the current epidemic stage in the country, expected transmissivity after lifting social distancing control, and the demographic profiles of the unvaccinated population. For example, low- and middle-income countries with high incidence and low vaccination coverage, the impact of vaccination and testing to reduce epidemic burden may be greater than our findings; high income countries with low incidence and high vaccination coverage, the impact of social distancing to reduce epidemic burden may be greater than our findings. As the impact of one particular or a set of strategies may differ by settings specific conditions as well as changing epidemic stages over time, our model may be utilised in various settings and repeated times to promote social consensus on policy decision and efficient resource allocation.

Our study has some limitations. First, our model did not consider possible reinfections and breakthrough infections among the vaccinated population due to a lack of data/data limitations. If we account for these factors, the level of the potential epidemic surge may be greater than our estimates. However, the impact on ICU demand may depend on the efficacy of the vaccine or available treatments (e.g., molnupiravir and paxlovid) to reduce disease progression.30 Second, it is difficult to assess the impact of the social distancing policies on reducing transmission levels (Rt). The current and future adherence to social distancing measures and their effects are subject to not only when and how long they are implemented but also who is targeted and where. Studies have shown a drop in the effects of social distancing over time with increasing behavioural fatigue and social/economic pressure. Moreover, the introduction of variants with increased transmissibility may require more stringent lockdown policies with stricter and longer social distancing to control the epidemic, especially in a setting with low vaccination coverage and constrained health systems capacity/access. As age-specific empirical data (e.g., different infectivity/transmissibility, progression, case fatality rates by age groups) become more available, future studies should explore the optimal targeting intervention strategies by age groups to minimize the extent of necessary restriction and achieve the best resource utilization. Third, the future epidemic resurgence will also be influenced by other factors such as herd immunity and immune waning effects.31 As seroprevalence data is yet to be released, the exact proportion of the undetected exposed population, their contribution to the transmission in the community, and their subclinical progression to natural recovery remain to be established. However, these factors may influence the overall susceptibility and herd immunity. While herd immunity along with high vaccination coverage may mitigate the potential epidemic resurgence, immune escape of new variants or waning immune effects against existing variants over time may collectively balance each other. The model can be calibrated in the future with real-time forecasting efforts and serological surveys to better reflect the COVID-19 prevalence and incidence within South Korea. Fourth, we did not consider other potential synergistic or counterbalancing effects when evaluating the parameters and combination scenarios. For example, the level of vaccine coverage may influence the susceptibility or progression rates in the population over time, influencing the infection levels and ICU demand. Similarly, the impact of testing (i.e., the number needs to test positivity) may differ with changing prevalence levels in society. In addition, strict countermeasures may reduce the testing rate itself or the testing effects, changing the population’s overall susceptibility. Finally, the model output may be heterogeneous due to population migration, mobility, and density; epidemic stage; and health system capacity. Future studies can further explore such heterogeneity across populations and regions and guide optimal control policies by setting specific conditions.

Despite these limitations, our model is the first compartment model to evaluate the relative intervention impact in South Korea based on detailed surveillance data obtained from the KDCA and other publicly available resources. By incorporating both the natural history of the COVID-19 infection and health system performance features, such as diagnostic performance and vaccination uptake and efficacy, this model comprehensively quantified the extent of uncertainty to examine competing policy options and guide an optimal exit strategy under possible epidemic and intervention scenarios.

In conclusion, this modelling analysis explored different intervention scenarios for an optimal COVID-19 exit strategy to reduce necessary social distancing measures and avoid epidemic surge in the future. We considered comprehensive intervention scenarios with different levels/duration of social distancing and testing intensity and vaccination rate under uncertain epidemic conditions with new emerging variants after lifting social distancing restrictions. Our findings suggest four or more months of social distancing control that can reduce 50% or more personal contact can effectively mitigate potential epidemic resurgence. In addition, increasing the testing intensity and vaccination rate by 20% may help reduce the level and duration of mandatory social distancing to prevent the epidemic resurgence and mitigate social and economic damage.

Supplementary Data 1

Model details

Supplementary Data 2

Multivariate uncertainty analyses

Supplementary Table 1

Model parameters, input values, and governing equations

Supplementary Table 2

Input parameters and ranges used in sensitivity analyses

Supplementary Fig. 1

Estimating key rate parameters. Rt from EpiEstim. We used EpiEstim (“estimate_R” function)1 to estimate time varying Rt based on the daily reported coronavirus disease 2019 cases in South Korea from 6 June to 21 November 2021 and assumed serial interval distribution (mean: 6, standard deviation: 2). The estimated Rt mean was 1.09 (median 1.04, interquartile range: 0.97–1.19), which was used to fit the observed data between 1 July to 21 November 2021. We calculated transmissibility based on Average Rt × Recovery Rate (1/14) × (1/contact) and assumed 1.7 times increased transmissibility from 17 October to 20 November 2021 by dominant delta variant influence and assumed three times increased transmissibility from 21 November 2021 to 31 December 2022 by the emerging new variant (e.g., Omicron) influence.

Supplementary Fig. 2

Estimating testing rate. The testing data represented information on the total number of tests performed per day, which was gradually increased since 2020. We estimated about average daily 35,000 (interquartile range 23,000–45,000) tests conducted in 2021 in South Korea. Since these tests were performed on both susceptible and exposed population, we assumed the average daily testing rate to be 0.14 among undetected asymptomatic population to fit to the observed incidence data, along with the current vaccination and transmissibility parameters. We then applied ±20% variation for the future testing rate from 21 November 2021 to 30 December 2022, for the sensitivity analyses.

Supplementary Fig. 3

Estimating age specific vaccination rates. We used information on COVID-19 vaccines administered between 6 March and 21 November 2021 from non-publicly available Korea Disease Control and Prevention Agency data. The vaccination data contained information on the weekly number of COVID-19 vaccines administered by type (AstraZeneca, Johnson & Johnson, Pfizer/Moderna) and order (1st, 2nd, and 3rd dose) by age group of vaccinated individuals (17 or younger, 18–19, 20–29, 30–39, 40–49, 50–59, 60–69, 70–74, 75–79, 80 or older years). We recategorized the 10 age groups into four (0–19, 20–39, 40–59, 60+ years). We used the respective vaccine efficacy for AstraZeneca, Johnson & Johnson, Pfizer/Moderna and assumed reduced vaccine efficacy by potential immune escape of the delta variant as shown (Table 1). We then combined the respective number of administered vaccinations by type and dose with the empirically observed proportion of the delta variants among all tested samples, to estimate variant and age specific effective vaccination rates per week. We then transformed the weekly rate into daily rate and calculated the average number of effective vaccinations administered per day by age group, and fitted them to the observed data (base case model) between 1 July to 20 November 2021 (9,787 for ages 0–19; 53,998 for ages 20–39; 73,070 for ages 40–59; 29,964 for ages 60+). In terms of the future vaccination rates, we assumed 3, 0.5, 0.5, and 0.1 times greater rates for the 0–19, 20–39, 40–59, and 60+ years age groups between 21 November to 30 December 2021 (29,361 for ages 0–19; 26,999 for ages 20–39; 36,335 for ages 40–59; 2,996 for ages 60+), 2, 0.3,0.2, 0.01 times greater rates for ages 0–19, 20–39, 40–59, and 60+ groups between 1 January 2022 to 30 June 2022 (19,574 for ages 0–19; 16,199 for ages 20–39; 14,614 for ages 40–59; 300 for ages 60+) and 0.2, 0.02, 0.02, 0.002 times (1,957 for ages 0–19; 1,079 for ages 20–39; 1,461 for ages 40–59; 60 for ages 60+) greater rates for ages 0–19, 20–39, 40–59, and 60+ between 1 July 2022 to 30 December 2022 based on the national vaccine priority and target strategies. We calibrated the future vaccination rates to satisfy the condition of an effective vaccination coverage of > 70% for each population group by December 2022. In our model, we assumed a prompt and lifelong vaccination immunity to be acquired from the day of immunization, given the vaccine effectiveness, by dose and by type of COVID-19 vaccine.

Supplementary Fig. 4

Partial rank correlation coefficient between input parameters and the COVID-19 incidence. We also computed PRCC4 describing the correlation between each input parameter and model output (i.e., the projected COVID-19 incidence in Scenario 1 as base case), adjusting for values of other input parameters. The PRCC ranked each parameter and output by the magnitude and measured the sensitivity of an output variable to each parameter. PRCC values range from −1 (perfect negative correlation between input parameter and model output) to +1 (perfect positive correlation), with values of 0 indicating no correlation. According to this finding, we conducted the multi-way sensitivity analysis based on the identified top three drivers (transmissibility, vaccination rate, testing rate) of the incidence outcome in Fig. 4.

Supplementary References