Article Text
Abstract
Background Excess mortality during the COVID-19 pandemic provides a comprehensive measure of disease burden, and its local variation highlights regional health inequalities. We investigated local excess mortality in 2020 and its determinants at the community level.
Methods We collected data from 250 districts in South Korea, including monthly all-cause mortality for 2015–2020 and community characteristics from 2019. Excess mortality rate was defined as the difference between observed and expected mortality rates. A Seasonal Autoregressive Integrated Moving Average model was applied to predict the expected rates for each district. Penalized regression methods were used to derive relevant community predictors of excess mortality based on the elastic net.
Results In 2020, South Korea exhibited significant variation in excess mortality rates across 250 districts, ranging from no excess deaths in 46 districts to more than 100 excess deaths per 100 000 residents in 30 districts. Economic status or the number of medical centres in the community did not correlate with excess mortality rates. The risk was higher in ageing, remote communities with limited cultural and sports infrastructure, a higher density of welfare facilities, and a higher prevalence of hypertension. Physical distancing policies and active social engagement in voluntary activities protected from excess mortality.
Conclusion Substantial regional disparities in excess mortality existed within South Korea during the early stages of COVID-19 pandemic. Weaker segments of the community were more vulnerable. Local governments should refine their preparedness for future novel infectious disease outbreaks, considering community circumstances.
- COVID-19
- Excess mortality
- Health inequalities
Data availability statement
Data are available in a public, open access repository. Data are available in public, open-access repositories, including the KOrean Statistical Information Service (KOSIS, https://kosis.kr), the MicroData Integrated Service (MDIS, https://mdis.kostat.go.kr), and the official website of the Korean Community Health Survey (CHS, https://chs.kdca.go.kr). Data from MDIS and CHS are free to access but require submitting a requisition form and undergoing an approval process on their respective websites.
This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.
Statistics from Altmetric.com
WHAT IS ALREADY KNOWN ON THIS TOPIC
South Korea had relatively few confirmed COVID-19 cases or deaths as a result of prompt and robust control measures during the COVID-19 pandemic.
WHAT THIS STUDY ADDS
Even a country such as South Korea, which had satisfactory average outcomes regarding COVID-19 control, encountered substantial regional health inequalities in excess mortality.
Pre-existing vulnerabilities in the community healthcare system were correlated with increased excess mortality.
HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY
Analysing the excess mortality with small regional units shed light on the unknown regional health inequalities during the pandemic.
Introduction
The COVID-19 pandemic, caused by the novel SARS-CoV-2 coronavirus, has had a significant impact on population health and society, varying greatly by region. A community’s vulnerability and resilience to the pandemic are determined by its physical and socioeconomic factors, further exacerbating regional health disparities.1–3 There is a pressing need to investigate these regional disparities in the context of COVID-19 pandemic.
Excess mortality serves as a valuable indicator that offers a comprehensive perspective on the short-term and long-term consequences of the COVID-19 pandemic.4–6 Excess mortality refers to deaths exceeding the expected levels based on past trends during a given period. It has been widely used to assess the nature, extent and impact of humanitarian emergencies and the effectiveness of interventions.7 Excess mortality related to the COVID-19 pandemic includes all-cause deaths directly or indirectly attributable to the virus. Indirect deaths refer to deaths arising from healthcare system overload, disruptions to care continuity, treatment delays or social restrictions during the pandemic.6 8 9
According to WHO estimates, there were 14.83 million excess deaths (95.12 per 100 000 individuals) across all 194 WHO member states during 2020 and 2021, a 13.16% increase over expected values without the pandemic.10 South Korea, a high-income Asia Pacific country, consistently demonstrated low rates of excess mortality in global comparisons. Estimates for 2020 and 2021 indicated 5620 COVID-19 deaths (5.4 per 100 000) and 4630 excess deaths (4.4 per 100 000) in South Korea.11
The pre-existing vulnerabilities synergistically amplified the impact of COVID-19, indicating a syndemic phenomenon.3 The community’s social, physical and economic conditions correlate with diverse pandemic impacts, further deepening regional health inequalities.12 Disadvantaged areas are disproportionately affected by COVID-19.13 The built environment, including public transportation, green spaces and pedestrian facilities, influences daily life routines and determines a broad range of health parameters.14 Publicly used facilities, such as religious places, shopping malls, movie theatres and public bathhouses, were the main sources of clustered outbreaks in South Korea.15 As a social process within a community, social relations are as crucial for crisis preparedness and response as infrastructure and economic resources.16 Social capital encompasses trust in others, reciprocity norms, social networks and civic engagement, which facilitate collective action for mutual benefit and improve societal efficiency.17 18 However, physical distancing policies aimed at curbing the spread of COVID-19 led to the disruption of social networks.19
To the best of our knowledge, previous research on excess mortality during the pandemic has focused on the national averages of various countries for global comparison and ranking. However, relying solely on the average values provides only a partial view of the pandemic, concealing disparities and equity issues within a nation. Even if a country reports favourable average outcomes, delving into smaller geographical units can reveal valuable insights into the pandemic’s impact. Furthermore, there is a lack of conclusive results regarding the impact of community determinants, which makes it challenging to formulate region-specific strategies.
In this study, we investigated regional disparities in excess mortality at the subnational level during the COVID-19 pandemic, along with the contributing factors. Our specific objectives were to estimate excess mortality in the 250 South Korean districts for the year 2020 and to investigate whether the excess mortality resulted from confirmed COVID-19 deaths or indirect deaths. Assigning the early stage of the pandemic as the study period will assist in assessing a capacity of a country to manage the novel shock. Furthermore, we identified relevant community predictors of regional excess mortality, considering both physical and socioeconomic factors within each community. Finally, we demonstrated the characteristics of communities vulnerable to the pandemic and proposed proactive strategies to address the needs of those areas.
Methods
Study area
South Korea comprises administrative regional units consisting of 17 cities and provinces, which are further subdivided into 250 districts. The dataset was organised and the variables were estimated according to these 250 districts.
Outcome variable
The excess mortality rate was defined as the observed all-cause mortality rate minus the expected mortality rate in 2020, based on the mortality trends of the previous five years (2015–2019). We used monthly data on all-cause mortality cases and population size by district from January 2015 to December 2020. These data were obtained from Vital Statistics20 and Population Statistics Based on Resident Registration21 via the KOrean Statistical Information Service (KOSIS).
To estimate the expected mortality rates, time-series techniques allowed us to precisely calculate the expected value, considering temporal trends, seasonality and other residual patterns22 ; therefore, the Seasonal Autoregressive Integrated Moving Average (SARIMA) was adopted. We identified the optimal SARIMA model parameters for each district to accommodate variable mortality patterns across regions using the R function, auto.arima.23 The fitted parameters based on the lowest Akaike’s information criterion with correction for small sample size (AICc) for each regional model are presented in online supplemental appendix A, Table 1.
Supplemental material
COVID-19-related variables
Data on COVID-19 confirmed deaths, sourced from the 2020 Cause of Death Statistics in Korea,24 can be accessed by submitting a request form to the MicroData Integrated Service (MDIS). The number of COVID-19 confirmed cases by city in 2020 was obtained from a press release by the Korea Disease Control and Prevention Agency (KDCA).25 Information regarding the levels of social restriction policies was collected from press releases issued during the pandemic by the Korean Ministry of Health and Welfare.26
Community determinants
Community characteristics in 2019 were collected as determinants of excess mortality. Data were retrieved from the KOSIS and MDIS repositories and the 2019 Community Health Survey (CHS),27 a nationwide survey conducted annually by the KDCA. We applied the CHS survey weights to calculate the representative prevalence rates by district. These determinants comprised five domains: (1) Residents’ health status included the proportion of individuals in each district with hypertension, diabetes or depression, and the proportion experiencing unmet medical needs, based on CHS data. (2) Healthcare resources were assessed based on the number of doctors, essential clinics, tertiary hospitals, nursing hospitals or social welfare facilities per 100 000 residents in each district. (3) Built environment features related to social restriction policies indicated the facilities with a high risk of viral transmission (eg, bars, singing rooms, bathhouses, restaurants, and cafés), health-promotion facilities (eg, cultural and sports facilities), parks and road areas. (4) Social capital indicators of trust, social networks, and social participation were estimated as the proportion of CHS survey respondents who answered the ‘yes’ option. (5) Socioeconomic characteristics included the ageing index, proportion of single-person households, employment rate, gross regional domestic product (GRDP) per capita, urbanisation rate and population density. Details of the variable descriptions and data sources are provided in online supplemental appendix C.
Statistical analysis
To evaluate the relationship between regional excess mortality rates and COVID-19-related events in cities and provinces, we plotted a smoothed curve using locally weighted regression (Loess) with optimal span values based on the lowest AICc. The correlation coefficients for the excess mortality rates and COVID-19-confirmed death rates by region are presented in online supplemental appendix B.
To select relevant predictors of excess mortality rates in 2020, we used three penalized regression methods: elastic net, lasso and ridge regression. These methods shrink the predictor dimension by imposing constraints on coefficient magnitudes through a tuning parameter (λ). Elastic net and lasso eliminate certain predictors by setting their coefficients to 0.28 Before applying the methods, we deleted four extreme outliers that exceeded lower bound (Q1−3×IQR) and upper bound (Q3+3×IQR) of the excess mortality rates from the dataset of 250 districts. We randomly split the total dataset (N=246) into a training set (70%) and a test set (30% of observations). Using the training set, we determined the optimal tuning parameter (minimum value of lambda, λ) for each method, ridge regression, lasso and elastic net, through 10-fold cross-validation. Subsequently, we evaluated the prediction performance of the three methods using the test set and selected the method with the smallest mean squared error (MSE). Finally, we applied the chosen method with the tuning parameter to the entire dataset and interpreted the selected predictors of regional excess mortality. The data preparation and analysis processes are detailed in online supplemental appendix D.
SAS V.9.4 was used for data preparation and descriptive results while R V.4.1.3 was used to present geographical distributions and analyse penalized regression methods.
Results
Estimates of regional excess mortality rates in South Korea
The predicted excess mortality rates in 2020 at the county level and across 17 cities and provinces are shown in table 1. The rates were estimated based on excess death cases from all causes per 100 000 people in each region. The national average did not increase significantly, with only 4.1 excess mortality rate. However, substantial regional disparities were observed, with excess mortality being higher in the provinces compared with cities. The Jellanam-do province reported the highest numbers of both observed and predicted deaths, with 42.1 excess mortality rate. Conversely, the Incheon, Daejeon, Gyeonggi-do, Jeollabuk-do and Jeju-do regions had excess deaths below the expected levels.
Figure 1 illustrates the regional distribution of excess mortality based on administrative borders. The nine provinces showed greater regional variation between districts compared with the eight cities (figure 1). Across 250 districts, the local excess mortality rates ranged from 10.8 in the 1st quartile to 62.7 in the 3rd quartile (online supplemental appendix A. Table 2). In 46 districts, the observed death rates remained below the expected values, with no excess mortality. In 127 districts, excess mortality rates were less than 50 while rates in 47 districts ranged from 50 to 100. Finally, 30 districts had rates exceeding 100 (online supplemental appendix A. Table 3).
Correlation between excess mortality rates and COVID-19
Figure 2 presents a smoothed curve depicting the relationship between COVID-19 and excess mortality, along with the population size of each city. The correlation between COVID-19 confirmed cases or deaths per 100 000 and excess mortality rates was not linear. Cities with high excess mortality rates (>20 excess cases per 100 000) were likely to have lower rates of COVID-19 confirmed cases and deaths per 100 000. The correlation coefficient (r) did not indicate statistical significance (p=0.769) at city and provincial levels (online supplemental appendix B. Table 4). With respect to social distancing policies, the three metropolitan cities in which these policies were rigorously enforced had fewer excess deaths per 100 000 people. However, in cities with medium or small populations, stricter restrictive measures were associated with higher excess mortality rates.
Predictors of excess mortality rates at the community level
Figure 3 shows the outcomes of penalized regression methods for identifying community predictors of excess mortality rates. Among the three methods, the elastic net was selected and fitted to the entire dataset because it indicated the smallest prediction error in the test set (test MSE=0.738). According to the elastic net result, 8 of 23 variables were selected as predictors. Communities with a high proportion of elderly residents or hypertensive patients were more likely to experience excess mortality. The number of doctors or hospitals did not emerge as a predictor. Instead, a higher number of nursing hospitals and residential social welfare facilities was associated with higher excess mortality. Increased social capital due to involvement in social and voluntary activities and social distancing policies in South Korea were effective for reducing excess mortality. Regarding built environment, facilities that promote healthy behaviours, including sports services and cultural infrastructure (ie, healthy facilities), were associated with a reduced risk for excess mortality. As cities become more urbanised, excess mortality tends to decrease. However, economic indicators, such as employment rate and GRDP per capita, were not identified as predictors.
Discussion
Local variation in excess mortality highlights the true impact of the COVID-19 pandemic. This study demonstrated significant regional disparities in excess mortality, ranging from 10.8 (Q1) to 62.7 (Q3) per 100 000 in 250 districts across South Korea. This can be partly attributed to the physical and social environments within communities. Rural and ageing areas, along with the presence of nursing hospital and social welfare facilities, and prevalence of hypertension, were associated with a higher risk for excess mortality. Both the reduction of physical contact and enhancement of social capital from civil society could protect against excess mortality.
Our results corroborate the low average of South Korea (only 4.1 excess deaths per 100 000) during the early stages of the pandemic, which is consistent with previous studies,11 22 29 30 demonstrating the effectiveness of the country’s social safety net. The national health insurance system in South Korea provides universal access to high-quality, low-cost medical services, reducing barriers to healthcare utilisation regardless of economic or employment status. In this study, factors related to economic status (eg, GRDP and employment rate) or healthcare utilisation (eg, unmet medical needs, number of doctors, essential clinics and tertiary hospitals) did not contribute to excess mortality.
Furthermore, we demonstrated significant heterogeneity among communities regarding excess mortality. We identified geographical concentrations of excess mortality in provincial (rural) areas. These localised excess mortality rates did not align with the rates of confirmed COVID-19 cases or deaths in those regions. This suggests that the increase in excess mortality is partly attributed to indirect deaths rather than direct COVID-19 confirmed deaths.
The predictors of excess mortality, identified using the elastic net method, included four primary aspects of focus. First, communities with higher proportions of elderly residents, lower levels of urbanisation or a higher prevalence of hypertension had a greater risk for excess mortality during the COVID-19 pandemic. This aligns with the persistent challenge of ‘rural and medically underserved areas’ in South Korea,31 suggesting that deprived regions were more severely affected by the pandemic.12 13 Meanwhile, this finding contrasts with the results of a study conducted in Stockholm, Sweden, which showed higher excess mortality in areas with younger populations during COVID-19 outbreaks.32
Second, there was a risk associated with nursing hospitals and residential social welfare facilities, consistent with findings from other countries. In England, residents of care and nursing homes faced a mortality risk approximately 10 times higher during the first wave of the 2020 pandemic compared with residents of similar ages in private homes in prepandemic periods.33 The elderly in long-term care facilities were vulnerable to both direct and indirect impacts of COVID-19.34 These findings suggest that the pandemic posed a high risk to patients in facilities where vulnerable people with restricted mobility gathered.
Third, our findings highlight the importance of both physical distancing policies and social capital when encountering novel infectious diseases. These seemingly contrasting findings are supported by previous studies. Latour et al 35 conducted simulations using Swedish mortality data and asserted that implementing a lockdown policy early in the COVID-19 outbreak could have saved more lives.35 However, stay-at-home policies have been associated with increased social isolation.36 Individuals who are socially isolated and experience loneliness, often with lower socioeconomic status, unhealthy behaviours and mental health issues, are at an increased risk of all-cause mortality.37 Striking a balance between physical distancing and social connections has become a crucial aspect of navigating a novel pandemic.
Finally, our results reveal the importance of the built environment in mitigating excess mortality. In our analysis, community facilities that provide sports services and cultural amenities were protective during the COVID-19 pandemic. These community structures may enhance physical activity and indirectly promote social and cultural capital, benefiting the health and well-being of residents.
These findings from South Korea have broader global implications. This study shows that even in countries with satisfactory national averages, significant regional disparities can exist with substantial variation across communities. In other countries, regional health inequalities within the nation may be even more pronounced than the reported figures suggest. Furthermore, this study suggests that it is the community’s pre-existing vulnerabilities rather than the viral infections alone that contribute to the increase in excess mortality. These underlying risks result in more critical and distal health consequences including deaths by disproportionately exposing individuals across sequential stages ranging from infection and access to care to economic hardship and recovery.13 As noted by Marmot et al 12 efforts should be made to ‘build back fairer’ against COVID-19 harms and reduce regional health inequalities by allocating more resources to disadvantaged areas that were damaged before and during the pandemic.12
This study had several methodological strengths. We estimated local excess mortality for 250 districts by fitting the SARIMA model to each district. Few previous studies have aimed to uncover disparities in excess mortality with a focus on small regional units during the COVID-19 pandemic. The elastic net had the smallest prediction error in this study. It sparsely selects predictors based on a penalty term, ensuring high predictability and interpretability.28 This method has the advantage of identifying crucial determinants among numerous explanatory variables and enables the establishment of priorities in response to future novel pandemics.
Our study also had several limitations. First, when estimating expected deaths, a trade-off emerges between the advantages and disadvantages of the forecast methods. The P-score, a widely used metric, establishes the previous 5-year average for weekly data as the baseline for normal deaths, ensuring comparability and interpretability.38 Conversely, the SARIMA method, we used, requires a parameter calibration process to predict the expected value, leading to increased complexity and difficulties in interpretation. Second, the estimated excess mortality rates in this study were not adjusted for or stratified by the sex and age structure of each district. Although a study in South Korea39 reported no significant sex differences in excess mortality patterns for 2020, further studies with disaggregated data by sex, age and local region will provide insights into the high-risk segments of excess mortality. Finally, caution is warranted when affirming that the excess mortality in South Korea in 2020 resulted solely from indirect COVID-19 deaths. Low COVID-19 death rates can be attributed to the country’s testing, diagnosis and reporting capabilities. These systematic biases, stemming from underdetected or unreported data, may lead to an underestimation of the risk posed by a novel pandemic.4 40 Hence, strengthening the country’s death registration system and investigating the causes of death contributing to excess mortality are crucial for understanding the pandemic and preparing for the future.
Conclusions
Considerable regional disparities in excess mortality have emerged despite satisfactory national average COVID-19 statistics in South Korea. Given that the Achilles’ heel of the community environment becomes a fatal flaw when faced with a novel pandemic, tailored countermeasures by local governments against a syndemic are imperative.
Data availability statement
Data are available in a public, open access repository. Data are available in public, open-access repositories, including the KOrean Statistical Information Service (KOSIS, https://kosis.kr), the MicroData Integrated Service (MDIS, https://mdis.kostat.go.kr), and the official website of the Korean Community Health Survey (CHS, https://chs.kdca.go.kr). Data from MDIS and CHS are free to access but require submitting a requisition form and undergoing an approval process on their respective websites.
Ethics statements
Patient consent for publication
Acknowledgments
This research was presented at the 3rd Joint Academic Conference of 5 Korean Societies to Identify and Solve the Causes of Regional Health Disparities on 27 October 2023. The conference was funded by the Korea Disease Control and Prevention Agency.
References
Supplementary materials
Supplementary Data
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Footnotes
Collaborators Not applicable.
Contributors EK and S-IC conceptualised and designed the study. EK conducted data curation, analysis and visualisation of the results and wrote the original draft. WL reviewed the methodology and contributed to the interpretation of the findings. All authors contributed to writing, reviewing and editing and approved the final version. S-IC is the guarantor for the study.
Funding This work was supported by the 2021 Research Programme on Facilitating Activities to Identify and Solve the Cause of Regional Health Disparities (grant number: ISSN 2733-5488), funded by the Korea Disease Control and Prevention Agency.
Disclaimer The funder was not involved in the study design; the collection, analysis and interpretation of data; the writing of the report; or the decision regarding publication.
Map disclaimer The inclusion of any map (including the depiction of any boundaries therein), or of any geographic or locational reference, does not imply the expression of any opinion whatsoever on the part of BMJ concerning the legal status of any country, territory, jurisdiction or area or of its authorities. Any such expression remains solely that of the relevant source and is not endorsed by BMJ. Maps are provided without any warranty of any kind, either express or implied.
Competing interests None declared.
Provenance and peer review Not commissioned; externally peer reviewed.
Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.