COVID-19 has corroborated insights gained from SARS, H1N1 influenza, and MERS pandemics showing socioeconomically disadvantaged populations are more severely impacted from pandemics [1
]. The reported gap between the COVID-19 rates of the most and least advantaged populations [5
] present a potential for reducing the outbreak through targeted interventions. Unlike non-modifiable factors [6
], such as populations with genetic predispositions, population vulnerability to the infectious disease outbreaks can be remediated through targeted interventions. Population vulnerability to respiratory infectious diseases is characterized by multiple interrelated factors, such as family income, education, employment status, health behaviour, healthcare access and other area-health indicators [8
]. Hence, identifying key socio-economic determinants for COVID-19 and mapping the vulnerable locales will enable policy makers to target specific modifiable factors in high-risk areas.
Among various approaches to disentangle how socio-economic status (SES) impacts health, Coleman’s social theory has been regarded as exceedingly useful because of its treatment of SES beyond access to material resources but also a function of social and human capital that ‘uniquely locate the individual’s status in the social structure
’. Blumenshine furthered the understanding by illustrating the mechanistic pathways between the socio-economic position and health disparity, as the underlying socioeconomic determinants of individuals can determine their likelihood of being exposed to the pandemic virus, contracting disease and timely and effective treatment after the disease developed [9
Although prior studies provided COVID-19 risk factors, none identified COVID-19-vulnerable locales associated with SES and COVID-19-specific epidemiological factors with acceptable generalizability and methodological capacity. Current COVID-19 studies relying on health disparity measures use arbitrary SES variables based on researchers’ preferences, irrespective of their COVID-19 relevance. Consequently, the SES measures across these studies are incomparable, limiting their usefulness. To address this, we integrated Coleman’s Social Theory and Blumenshine’s mechanistic framework (Figure 1
), which formulates a universal SES definition and SES indicator selection mechanistically/causally relevant to the COVID-19 health outcome. This approach can inform public health interventions to alleviate SES factor-related COVID-19 risk.
Since SES and epidemiological data cover an expanse of highly intercorrelated variables, the composite SES index, derived from multiple unique SES variables help garner the most explanatory information from the contributing indicators. A single universal composite measure, like the commonly-used area-deprivation index [10
], is limited to controlling SES effects as confounders, but not when the study’s goal is assessing the effect of multiple SES determinants on health outcome (e.g., COVID-19 in this study). Therefore, a multiple composite SES index approach helps quantify each composite SES index’s effect on COVID-19 [11
Thus far, one study used multi-scale Geographically Weighted Regression (GWR) to map the US COVID-19 incidence rate, while accounting for selected SES variables (median household income, income inequality, percentage of nurse practitioners, and black female population) [12
]. Since multi-scale GWR doesn’t fit a beta distribution typical for infectious disease rates [13
], we recommend Geographically Weighted Negative Binomial Regression (GWNBR) to improve methodological accuracy. GWNBR directly uses discrete count data without further transformation, and is robust in the overdispersion, spatial/temporal clustering and false-positives [14
Globally, the COVID-19 pandemic emerged in waves with country-specific mitigation strategies producing sharp declines. To improve public health interventions by precision targeting of high-risk locales, this study identified key SES and epidemiological risk determinants and their geographic distribution. We chose South Korea because its COVID-19 incidence data presented extremely high overdispersion, temporality, and spatial clustering, being more complex than typical infectious disease data. This allowed us to check our framework’s functionality to address the dramatic spatial and temporal dynamic of COVID-19 [16
]. Our study’s goals were to; (1) provide methodological framework for identifying COVID-19-vulnerable locales associated with SES and epidemiological determinants; and (2) operationalize the framework using South Korean data to demonstrate its value.
This study provided a methodology to map COVID-19 risk associated with multiple SES and pandemic-specific epidemiological factors with high geographical granularity. We refined a social theories-enhanced CBF model to guide our variable choices. This model warrants the potential variables identified from the COVID-19 risk factor literatures encompass the conceptual domains of (material, human, and capital) SES and plausible to cause differential exposure, susceptibility, and disease severity. The simultaneous influence of multiple SES determinants defines disease risk. By creating multiple composite SES indexes using PCA, this study clarified whether certain SES determinants independently contributed to the COVID-19 risk over and above other SES factors.
GWNBR created a continuous surface of relative COVID-19 risk for all 250 districts associated with area-health and socioeconomic determinants by the pandemic phases (Figure 4
). Our findings are consistent with individual and population-level studies that reported elevated COVID-19 risk associated with less healthcare access
], and education
], and more risky health behaviour
, specific comorbidities
], difficulty to social distancing
] and population mobility
]. Our study’s high internal validity was shown since the GNBR and GWNBR results agreed except for crowding
in the early phase.
Our approach captured statistically and noticeably high spatial variation by pandemic phases for all themes, consistent with the reported pattern of COVID-19 distribution in the country [22
]. Since its first confirmed case on January 20th, 2020, South Korea experienced two major outbreak waves in Daegu and Seoul, and the surrounding Gyeonggi-do province, respectively, in February (early phase) and May 2020 (late phase).
After the first confirmed case, the KCDC invoked four-level alert for the public’s emergency awareness (blue-attention, yellow-caution, orange-alert, red-serious) commensurate with the number of new confirmed cases [http://www.koreabiomed.com
]. The epidemic’s initial wave in Daegu, caused by the local church activities, triggered the country-wide directives of hospital-based isolation/quarantine, contact tracing followed with free testing and treatment, strengthening medical centres for rapid diagnostics, emergency medical responses, and treatment aids [22
]. These specific measures along with high public adherence to the school and business closures, personal hygiene, and social distancing significantly dropped the case counts by mid-March. The second wave erupted in May when non-essential businesses reopened [28
] in Seoul, which spread to its surrounding metropolises, Ulsan and Busan, and Gyeonggi-do.
The types of risk determinants changed over the pandemic phases
. Analysis stratified by periodic phases found that the initially high risk in the early period gradually decreased except healthcare access, health behaviour and crowding- associated risk, which increased in strength and concentrated in the capital and its surrounding provinces in the late phases (Figure 3
and Figure 4
). Risk reductions could be explained by the impact of effective control measures that lowered the risk associated with these determinants and drop in an effective reproduction number (Re), as the number of infection-susceptible people decreased over time [33
In the early phase, all health/SE themes were statistically significantly associated with COVID-19 incidence. As anticipated, there was no excess risk at Daegu and its surrounding areas since the abnormally high spike of COVID-19 cases was caused by local church’s activities [22
] without relevance to the local area’s social status.
The increase in health behaviour-associated risk is consistent with reports showing greater risk with poor emotional health [30
], smoking [34
], and obesity [1
]. Individual patient-level COVID-19 risk-factors analysis Lusignan et al.,2020 [1
] reported smoking was a protective factor. However, the authors warned that the low proportion of current smokers in their study sample (11.4%), resulted in a wide confidence interval of the reported odds ratio, 0.59 (0.42–0.83). This increases the uncertainty of their result. Greater COVID-19 risk among the people with lower education has been explained as a reduced awareness of disease risk and low-income to obtain education. This relationship was clearly seen in our results as higher education associated with lower COVID-19 incidence. The difficulty to social distancing in this study directly reflected the inability to afford unemployment possibly resulting in the reduced exertion of protective measures [30
]. The risks associated with social distancing and area-morbidity peaked in the study’s early phase appeared to reduce in the middle phase, and completely remediated in the study’s late phase.
In the middle phase, all of the previous risk factors except for risky health behaviours, population mobility, and crowding were high. The middle phase’s lessened risk associated with risky health behaviours, population mobility and crowding may reflect the impact of the Prime Minister’s declaration. This implemented active interventions for social distancing, community health education, testing with local contact, tracing, and hospital-based or self-isolation during March’s first weeks.
Notably, the late phase findings are consistent with the risk factors reported associated with the second wave in early May. During our study’s late phase, South Korea scaled up free testing and treatment through its existing health care centres [28
], which may have improved healthcare access a key measure for combating COVID-19. Our finding that healthcare access exerts a stronger protective effect in the late phase compared with the earlier phases supports this. Our findings of increased risk associated with risky health behaviours may have captured behavioural fatigue at a population-scale in response to the country’s multiple quarantine period extensions [37
] that likely were exacerbated by entertainment business re-openings (i.e., night clubs, karaoke) in early May. Elevated risks associated with increased crowding in the study’s late phase reflects the outbreak’s second wave, which occurred in South Korea’s most crowded region: Seoul and its surroundings (Figure 3
Spatial variation in the SES-related risk factors across the pandemic phases potentially reflect the geography-specific control measures and/or the differential public response to the measures.
GWNBR models revealed the pandemic phase-specific spatial variation for all health/SE themes except for population mobility which was not significant beyond the early phase. This may indicate that the effectiveness of the control measures varied over time potentially due to differential interventions or public response across the municipal districts. Our findings may also indicate a dynamic change in population vulnerability throughout the pandemic “a person not considered vulnerable at the outset of a pandemic can become vulnerable depending on the policy response” as a Lancet editorial stated [38
The factors increasing our recommended framework’s robustness include: (1) SES measurement and relationship conceptualization of the exposure (health/SE themes) and outcome (COVID-19 incidence) based on the refined conceptual framework; (2) joint use of conceptual and statistical modelling; (3) complementary use of global and local spatial statistics; and (4) stratified analysis by pandemic phases that enable us to capture the spatial variation over pandemic phases. However, this methodological framework relies on carefully collected country-specific data.
Our study is subject to ecological fallacy inherent to the study design. However, our empty hierarchical mixed model accounting for the individual and district-level data shows that 61% of the COVID-19 incidence distribution variation was explained by the district-level factors, leaving 39% of the variability for an explanation by individual factors.
We verified that the data estimation for Daegu city subparts did not affect the study results. The comparison of the intercept, standard error, relative risk and p
-value between the models with and without the estimated data showed that the intercept and standard error were diminished by 2.2% and by 9.2%, respectively in the models, including estimated data [each calculated by 100
(−12.29 − (−12.56))/−12.29 and 100
(1.88 − 2.95)/1.88, respectively]. A significance level change was observed for none of the model estimates, except for crowding.
-value changed from ~0.06 to ~0.04 when the estimated data were excluded. However, the crowding-associated
risk remains significant at p
= 0.1. Model details are provided in Table S4
. To assess the periodic trend in the relative COVID-19 risk associated with SES factors, we conducted stratified analyses by the early, middle, and late phases corresponding with 20 January–20 March, 21 March–15 April, and 16 April–1 July 2020.
Population-based COVID-19 studies are prone to response bias, which would not exist if everybody was tested. However, multiple factors determine testing coverage, therefore, the number of confirmed cases, such as easy access for testing and its accuracy [39
], contact tracing strategies [40
], under-testing of asymptomatic patients [41
]. Also, psychological factors, a fear of COVID-19 [42
], risk perception [43
], and stigma related testing avoidance [46
] impact testing rate. Potential bias in this study is expected to be low given South Korea’s anti-pandemic strategies. Importantly, the country’s COVID-19 relief programs supported with 15 billion Korea won, dedicated to support vulnerable populations may have reduced the potential testing disparity by socioeconomic status. All Koreans and foreigners were entitled to free testing and treatment, while testing access was more convenient through an extended number of rapid diagnostic centres and testing prompts through mobile phones. Contact tracing-based testing increases the likelihood of capturing asymptomatic cases. Korean tracing system has been reported as the global best practice lending to its advanced information technology system and data extensions through large consumer and healthcare databases (global positioning system, credit card transactions, closed-circuit television and medical facility use records) [47
]. Testing avoidance from fear of stigma [46
] would likely have affected the early period of the analysis, which strongly reflected the abnormally high spike of cases in Daegu city traced to the local church. The city has not disclosed the data with necessary granularity for a further investigation of this matter, as of writing. However, given these factors would likely result in under-estimation of confirmed cases, any bias in our results should be toward the null.
The demonstrated methodology guides to design of multiple-determinant targeted interventions and pinpoint high-risk locales to remediate the excess COVID-19 risk attributable to socioeconomic disadvantages. Overall, our work has demonstrated that the anti-pandemic measures taken by the South Korean government were effective.
The completely remediated risk associated with area-morbidity and difficulty to social distancing is likely to be explained by the country’s emergency relief programs that targeted vulnerable individuals with socioeconomic disadvantages: Foreign workers, homeless, poor urban residents, disabled people, and elderly. The assistance programs provided free testing, financial support, food assistance, health check-up visits, as they acknowledged excess hardship in adhering to social distancing rules because of inability to afford unemployment.
The observed overall protective effect of improved healthcare access and higher education in our study support the rationale behind the country’s primary anti-pandemic agenda to strengthen healthcare facilities for rapid diagnostic and therapeutic services, combined with actionable health promotion rules which reportedly gained high public compliance.
However, we found risky health behaviour was a persistent risk factor during both major outbreaks in Daegu and Seoul. Elevated crowding associated risk coincided with the Seoul outbreak, as anticipated.
Persistently high-risks associated with health behaviour and crowding, combined with the reduced protective effect of healthcare access and education in the study’s late phase may corroborate the finding that a prolonged pandemic induces adherence fatigue and lessened risk perception [31
South Korean public health interventions have been discussed in detail elsewhere [28
] and we endorse the country’s anti-pandemic interventions as guidance to international policymakers. The main highlights were: (1) Targeting vulnerable locales to COVID-19 and aiming to address multiple risk factors considering emergency relief programs to provide financial support, food assistance, health check-up visits; (2) implementing social distancing while assisting individuals with difficulty to social distancing. Social distancing measures may include school/business closures, hospital-based or self-isolated quarantine; (3) improving healthcare access for expanded testing and treatment with priority health services made available to the individual with extenuating medical conditions; (4) strengthening existing healthcare facilities and extending rapid diagnostic centres to enable easily accessible and free testing; (5) enhancing case identification capacity through information technology, such as contact tracing, mobile phone-based testing prompts and general risk alerts, rapid case-isolation by automated test result delivery to the testee’s mobile phones.
We emphasize the importance to anticipate adherence, behavioural, and mental fatigue over the course of a prolonged epidemic. In the latter phase of the epidemic, we recommend paying intensified attention to the urban and highly crowded areas to prevent a potential outbreak as well as promoting creative social networking solutions (drive-through services, virtual social events, telehealth, etc.) and ensuring emerging vaccine accessibility for the socially disadvantaged population [50
We intend that this framework can be replicable to both, international researchers and policymakers, in order to enable rapid pandemic responses. As socioeconomic disparity is a global problem, nationwide programs with an intensified focus on the vulnerable populations at excess risk to pandemic ensure the efficacy and efficiency of pandemic alleviation efforts.
Future research should assess the mortality, and mortality and incidence ratio as a crude surrogate for survival using the same study design and methodology. Understanding the impact of socio-economic and epidemiological risk factors on mortality compared with incidence would clarify the extent of the potentially preventable deaths through modifications of the assessed risk factors. The overall and spatial disparity between mortality, incidence, and mortality/incidence ratio would inform where intensified public health and intensified healthcare services are needed.