https://orcid.org/0000-0001-7472-3752
Abstract
Previous studies have reported simple differences in extreme temperature-related health risks by low socioeconomic status; however, few have examined in depth the double disparities in the socially marginalized people by age groups, sexes, disabilities, and causes of hospitalization. This study examined (i) the differences between heat- and cold-related risks on hospitalization between people who are eligible and non-eligible for the medical aid system in the national health insurance service system and (ii) differences between the heat- and cold-related risk and cost differences by specific subgroups in South Korea.
We collected population-based longitudinal cohort data from the National Health Insurance Service-National Health Insurance Database from 2010 to 2019. The data included all individuals who were eligible for the Korean Medical Aid (MA) system during the study period and we used their data on hospitalization through the emergency department (ED). As a control group, we collected age–sex–residential address-matched individuals who were not eligible for the MA system. We adopted a case-crossover design with a distributed lag nonlinear model to evaluate the excess risks and costs associated with cold and heat temperatures on hospital admissions via the emergency room department.
During the study period, 509 480 hospital admissions via the ED were recorded among 1 466 176 beneficiaries who were eligible for MA. Among the MA beneficiaries, the estimated risk for ED admission that was attributable to heat was 1.19 [95% confidence interval (CI): 1.14–1.24] and the risk that was attributable to cold temperature was 1.52 (95% CI: 1.43–1.61), which were both higher than those of the control groups that incorporated matched beneficiaries who were not eligible for MA. For both heat and cold, the difference between MA and non-MA was prominent in non-elderly populations, males, people with disabilities, and admissions with mental and cardiovascular diseases.
This study revealed the hypothesis that the differences between heat- and cold-related risks in the socially marginalized population existed and suggested that the disparities might also be disproportionate by socioeconomic and demographic statuses.
We investigated how extreme temperatures were associated with hospitalization, revealing double disparities between Medical Aid (MA) and non-MA populations and within MA beneficiaries.
The cold and heat risks were higher in MA individuals than in non-MA individuals, and the risk differences were generally more prominent in young and middle-aged (aged 0–64 years) populations, hospitalizations for cardiovascular and mental disorders, and people with disabilities than in the total population.
This study highlights the double disparities among MA beneficiaries, providing key insights for targeted public health interventions.
Introduction
Extreme temperatures have hazardous effects on human health and can even lead to death [1–5]. Previous studies have been conducted on whether there is health disparity for the poor (e.g. low-income or medical aid individuals) that is related to temperature [6–8]. Few studies have, however, investigated the double disparities in economically disadvantaged populations, which indicate that they can experience other disparities such as sex, age, and disability in addition to their economic disparities. This can generate synergism, with the nationwide population focusing on the multi-morbidity outcomes and medical costs associated with extreme temperature exposure.
It is important to understand the double disparities, especially when evaluating risks due to extreme temperatures. First, marginalized populations can be more likely to reside under poorly insulated conditions with inefficient heating or cooling systems [7–15]. Second, people who have outdoor jobs might be more frequently and intensively exposed to extreme temperatures for extended periods and, in general, this outdoor working group (included in blue-collar occupations) has relatively lower socioeconomic and health conditions compared with people with indoor working conditions [16]. The financial constraints also limit access to healthcare, delay treatment for temperature-related illnesses, and worsen health outcomes [17–19]. These combined risk factors can generate disproportionate vulnerabilities even among economically marginalized populations. Thus, it is crucial to carefully examine the “disparities in disparities” regarding health risks and costs for extreme temperatures.
In this study, we investigated the double disparities in health impacts of temperature exposure among beneficiaries of the Medical Aid (MA) program in South Korea, which is an underexplored subgroup of the economically marginalized population. With nationwide longitudinal cohorts of MA beneficiaries, we evaluated the excess risks and costs due to temperature exposures (heat and cold) and identified more vulnerable subgroups in MA beneficiaries.
Methods
This study was approved by the Institutional Review Board of Seoul National University, Seoul, South Korea (IRB number E2302/004–002).
Study population
This study used the 2010–2019 National Health Insurance Service-National Health Insurance Database (NHIS-NHID) of South Korea. We collected data for all beneficiaries whose type of insurance included members of the household who had been receiving MA from 2010 to 2019 as the study group. MA beneficiaries are eligible for the Korean MA system—one of the social security systems along with health insurance as a state-guaranteed public assistance system for low-income citizens who cannot sustain basic living conditions, very similar to Medicaid in the USA [20–21] (described in more detail in the Supplementary material in ‘Korean health security system’).
We defined the control group with matched data for people who were not eligible for the MA system (subscribers to the Korean National Health Insurance) with the Medical Aid beneficiaries stratified by sex, age, and residential area (‘Si-gun-gu’ district). The NHIS-NHID data contained information on socio-demographic characteristics (e.g. sex, district-level address, income percentile, and disability type) and hospital claims (e.g. admission date, diagnostic codes, and medical costs).
Temperature data
We obtained temperature data from the ERA-5 Land dataset—a climate reanalysis dataset from the Google Earth Engine (Supplementary Table S1; see online supplementary material for a color version of this table). We collected the daily mean temperature data over a 9-km spatial grid and aggregated them for each district (‘Si-gun-gu’) by averaging the modeled values at the grid cells with centroid points inside the boundary of that district: the range of the size of districts was 3.0–1820.6 km2 (median: 443.0 km2) [22–24]. The temperature percentile in each district during the entire study period was used as the exposure. Supplementary Fig. S1 (see online supplementary material for a color version of this figure) shows that our modeled temperature data showed a very high prediction accuracy (R2 = 0.97) compared with the monitoring stations.
Hospitalization data
Hospitalization data were collected from emergency department (ED) admissions at the NHIS-NHID, including the admission date, total medical cost, and main diagnosis code for the admission. The main diagnostic code is based on the 10th Revision of the International Statistical Classification of Diseases (ICD-10).
To examine double disparities among the study population, we defined subgroups by age, sex, disability status, and causes of ED hospitalization. We classified age into four groups: –18 years, 19–64 years, 65–84 years, and 85+ years of age. Sex was classified as male or female, and the disability group as people with disabilities and people without disabilities. In addition, we classified causes of ED hospitalization into four main diagnoses including cardiovascular diseases (ICD-10, I00-I00), genitourinary diseases (ICD-10, N00-N99), mental disorders (ICD-10, F01-F99), and respiratory diseases (ICD-10, J00-J99). Detailed information on the diagnostic codes is displayed in Supplementary Table S2 (see online supplementary material for a color version of this table).
Statistical modeling
We estimated the association between ED admissions and short-term exposure to ambient temperatures by using a conditional logistic regression based on a time-stratified case-crossover design [23–24]. In this study, participants served as their own controls and the risk inference was based on a comparison of the daily ambient temperatures on the case day versus the daily ambient temperatures on the control days [25]. Specifically, the case day was defined as the admission date for each ED visit and the control days were selected in the same year and month as the case day to bidirectionally control for seasonal and long-term trends [23]. We performed a stratified analysis to evaluate the association between extreme temperature and ED visits for MA and non-MA groups because of a computational issue: the NHIS internal server allocated limited memory size for each research area, thus the time-stratified datasets for the total cases (i.e. the dataset the included both MA cases and non-MA control groups) and several subgroups with a large number of ED visit cases could not be analysed by using case–control integrated data owing to the memory problem. Thus, we could not conduct conditional logistic regression models with an interaction term (an indicator variable for MA and non-MA individuals) and performed stratified analyses instead.
We then performed a distributed lag nonlinear model with a conditional logistic regression adjusting for holiday as a confounder [24, 26]. We modeled exposure–response functions by using a natural cubic spline with four degrees of freedom and lag-response functions by using a natural cubic spline with two knots placed at lag 1 d and lag 3 d on the lag up to 21 days. We set the 75th percentile of the temperature as a reference point for both MA and non-MA control groups to calculate the risk estimates of extreme heat and cold temperatures [27]. To show the appropriateness of the reference point selection, we tried to estimate the exposure–response curve by using the data that incorporated both groups; however, we could not perform it due to the limited computational memory size of the NHIS data analytic server. Thus, as an alternative method, we derived the minimum hospitalization temperature (MHT) as the reference temperature by randomly sampling 10% of the total data 100 times; as a result, the most frequently identified MHT was the 75th percentile, consistently with the previous studies [23, 27]. We specified the cold and hot temperatures by using the following temperature percentile values: extreme cold (1st percentile temperature) and extreme heat (99th percentile temperature) [23, 27, 28].
To express the risks of extreme temperatures, we used odds ratios (ORs), which were calculated as the odds at the defined temperatures (heat and cold) compared with those at the constant MHT (75th percentile temperature). To examine the differences in ORs between the MA and non-MA groups, we also estimated the relative odds ratios (RORs): . We calculated the 95% of confidence intervals (CIs) of the RORs by using the formula based on the independent assumption: . In addition, to examine the statistical evidence of the ROR estimates, we performed a univariate Wald test.
We estimated the excess medical costs that were attributable to temperature [23, 26], which can be calculated by using , where α is calculated by using the daily total healthcare cost of the ED admission and OR is the estimated value with each population [23, 26]. We then calculated the attributable costs per year to show the annual increase in medical costs that was attributable to temperature considering 100 000 persons/year. The excess medical costs that were attributable to temperatures were estimated for the levels of overall temperature, cold temperature (1st temperature percentile to MHT), and hot temperature (MHT to 99th temperature percentile) [23, 26]. The 95% empirical confidence intervals (eCIs) for the attributable ED admissions were determined by using 1000 Monte Carlo simulations [23, 26].
We performed sensitivity analyses by testing different model specifications. SAS (Enterprise v7.1, SAS Institute Inc, Cary, NC) was used for the data preprocessing of each cohort and R (version 3.3.3) was used for the statistical analyses with the packages: “clogit,” “tsModel,” and “dlnm.”
Results
During the study period, 509 480 hospital admissions through the ED were recorded from 1 January 2010 to 31 December 2019 among 1 466 176 beneficiaries who were eligible for the Korean MA system. With the yearly baseline values per 100 000 persons, MA beneficiaries showed a higher baseline ED admission rate (3474.9) and medical costs (US$6 547 700) than non-MA beneficiaries (baseline admission rate = 1299.1; baseline medical costs = US$3 196 300). Detailed information on the ambient temperatures is presented in Supplementary Table S3 and Supplementary Fig. S2 (see online supplementary material for color versions of the table and figure).
Table 1 shows the summary statistics of the admissions into EDs with the individual characteristics among the MA beneficiaries (case group) and matched non-MA beneficiaries (control group). Among the MA beneficiaries, 52.0% were female and 54.8% were elderly (≥65 years). Meanwhile, among beneficiaries who were not eligible for the MA system (matched cohort), 56.1% were female and 71.0% were elderly (≥65 years) (Table 1). Major causes of hospitalization for the MA and non-MA groups are reported in Supplementary Tables S4 and S5 (see online supplementary material for color versions of these tables) and they were not obviously different.
Total number of hospital admissions through the ED among MA beneficiaries (2010–19)
| Variable | Total number of ED admissions among MA beneficiaries (%) | Total number of ED admissions among non-MA beneficiaries (%) |
|---|---|---|
| Total number | 509 480 | 213 849 |
| Age (years) | ||
| –18 | 9139 (1.8) | 6445 (3.0) |
| 19–64 | 220 872 (43.4) | 55 502 (26.0) |
| 65–84 | 209 621 (41.1) | 115 043 (53.8) |
| 85+ | 69 848 (13.7) | 36 859 (17.2) |
| Sex | ||
| Male | 244 765 (48.0) | 93 977 (43.9) |
| Female | 264 715 (52.0) | 119 872 (56.1) |
| Disability status | ||
| People with disability | 266 583 (52.3) | 48 415 (22.6) |
| People without disability | 242 897 (47.7) | 165 434 (77.4) |
| Cause of visit | ||
| Cardiovascular | 164 562 (32.3) | 93 185 (43.6) |
| Genitourinary | 83 055 (16.3) | 42 828 (20.0) |
| Mental | 123 926 (24.3) | 8125 (3.8) |
| Respiratory | 137 937 (27.1) | 69 711 (32.6) |
| Variable | Total number of ED admissions among MA beneficiaries (%) | Total number of ED admissions among non-MA beneficiaries (%) |
|---|---|---|
| Total number | 509 480 | 213 849 |
| Age (years) | ||
| –18 | 9139 (1.8) | 6445 (3.0) |
| 19–64 | 220 872 (43.4) | 55 502 (26.0) |
| 65–84 | 209 621 (41.1) | 115 043 (53.8) |
| 85+ | 69 848 (13.7) | 36 859 (17.2) |
| Sex | ||
| Male | 244 765 (48.0) | 93 977 (43.9) |
| Female | 264 715 (52.0) | 119 872 (56.1) |
| Disability status | ||
| People with disability | 266 583 (52.3) | 48 415 (22.6) |
| People without disability | 242 897 (47.7) | 165 434 (77.4) |
| Cause of visit | ||
| Cardiovascular | 164 562 (32.3) | 93 185 (43.6) |
| Genitourinary | 83 055 (16.3) | 42 828 (20.0) |
| Mental | 123 926 (24.3) | 8125 (3.8) |
| Respiratory | 137 937 (27.1) | 69 711 (32.6) |
Total number of hospital admissions through the ED among MA beneficiaries (2010–19)
| Variable | Total number of ED admissions among MA beneficiaries (%) | Total number of ED admissions among non-MA beneficiaries (%) |
|---|---|---|
| Total number | 509 480 | 213 849 |
| Age (years) | ||
| –18 | 9139 (1.8) | 6445 (3.0) |
| 19–64 | 220 872 (43.4) | 55 502 (26.0) |
| 65–84 | 209 621 (41.1) | 115 043 (53.8) |
| 85+ | 69 848 (13.7) | 36 859 (17.2) |
| Sex | ||
| Male | 244 765 (48.0) | 93 977 (43.9) |
| Female | 264 715 (52.0) | 119 872 (56.1) |
| Disability status | ||
| People with disability | 266 583 (52.3) | 48 415 (22.6) |
| People without disability | 242 897 (47.7) | 165 434 (77.4) |
| Cause of visit | ||
| Cardiovascular | 164 562 (32.3) | 93 185 (43.6) |
| Genitourinary | 83 055 (16.3) | 42 828 (20.0) |
| Mental | 123 926 (24.3) | 8125 (3.8) |
| Respiratory | 137 937 (27.1) | 69 711 (32.6) |
| Variable | Total number of ED admissions among MA beneficiaries (%) | Total number of ED admissions among non-MA beneficiaries (%) |
|---|---|---|
| Total number | 509 480 | 213 849 |
| Age (years) | ||
| –18 | 9139 (1.8) | 6445 (3.0) |
| 19–64 | 220 872 (43.4) | 55 502 (26.0) |
| 65–84 | 209 621 (41.1) | 115 043 (53.8) |
| 85+ | 69 848 (13.7) | 36 859 (17.2) |
| Sex | ||
| Male | 244 765 (48.0) | 93 977 (43.9) |
| Female | 264 715 (52.0) | 119 872 (56.1) |
| Disability status | ||
| People with disability | 266 583 (52.3) | 48 415 (22.6) |
| People without disability | 242 897 (47.7) | 165 434 (77.4) |
| Cause of visit | ||
| Cardiovascular | 164 562 (32.3) | 93 185 (43.6) |
| Genitourinary | 83 055 (16.3) | 42 828 (20.0) |
| Mental | 123 926 (24.3) | 8125 (3.8) |
| Respiratory | 137 937 (27.1) | 69 711 (32.6) |
Different exposure–response curves, depending on MA eligibilities, are shown in Fig. 1. Across the overall temperature, the risk of temperature-related ED hospitalization was higher among beneficiaries who were eligible for the MA system than that among those who were not.
Overall cumulative lag temperature percentile versus ED admissions in people eligible for MA (blue) and people not eligible for MA (red). Data are ORs (95% CI; red line). The vertical dotted lines indicate the 1st, 10th, 90th, and 99th percentiles. The shaded regions represent 95% CIs. The vertical line shows the MHT (75th percentile).
Figure 2 shows the risks and excess costs due to extreme heat by MA eligibilities and subpopulations, concurrently. Considering the RORs in panel 2, the largest value in the risk of ED admission according to MA eligibility was observed among people aged ≤18 years (ROR = 1.60, 95% CI: 1.12–2.10). Stratified by causes of hospitalization, MA beneficiaries numbered the highest (OR = 1.43, 95% CI: 1.32–1.55) for mental disorders compared with non-MA beneficiaries (OR = 1.20, 95% CI: 0.89–1.63). Considering excess medical costs, people aged ≥85 years who were eligible for the MA system had the highest excess medical costs (US$25 300; 95% eCI: –16 500 to 63 200) per year per 100 000 persons/year, which was higher than the excess medical costs for non-MA beneficiaries. Further, based on the point estimates, the risk differences by MA eligibility were slightly more prominent in males, people with disabilities, and admissions with cardiovascular diseases.
Risk of admissions through ED visits and annual increase in the associated medical costs (US$1000) attributable to extreme heat. The first panel shows the OR when using a case-crossover design with a conditional logistic regression. The second panel shows the ROR that was calculated as the OR of persons eligible for MA divided by the estimates of persons not eligible for MA. Error bars show 95% CIs for estimates of risk. The third panel shows the annual increase in medical costs. It was calculated as α × (OR – 1)/OR, where α is the baseline daily medical costs from 2010 to 2019 per 100 000 person-years. α was calculated among each study group divided by the total person-years for beneficiaries and multiplied by 100 000.
As shown in Fig. 3, with cold temperatures, the relatively younger age group (aged 19–64 tears) had the highest risk difference for extreme cold, showing that the OR for MA beneficiaries was 1.65 (95% CI: 1.51–1.80), 1.28 (95% CI: 1.06–1.53) for non-MA beneficiaries, and the ROR for MA eligibility was 1.29 (95% CI: 1.09–1.50). Further, the cold OR among males who were eligible for the MA system (OR = 1.65, 95% CI: 1.52–1.80) was higher than for those who were not eligible (OR = 1.36, 95% CI: 1.18–1.56). Excess medical costs increased among the older group and showed higher excess costs among MA beneficiaries than among non-MA beneficiaries in each age group. The corresponding values to those in Figs 2 and 3 with statistical test results are reported in Supplementary Table S6 (see online supplementary material for a color version of this table). We also estimated ORs for moderate temperatures and detailed causes of hospitalizations (Supplementary Tables S8–S10; see online supplementary material for color versions of these tables) and the general trend (higher risks in MA people) was consistent in the detailed causes of hospitalization analyses.
Risk of admissions through ED visits and annual increase in the associated medical costs (US$1000) attributable to extreme cold. The first panel shows the OR when using a case-crossover design with a conditional logistic regression. The second panel shows the ROR that was calculated as the OR of persons eligible for MA divided by the estimates of persons not eligible for MA. Error bars show 95% CIs for estimates of risk. The third panel shows the annual increase in medical costs. It was calculated as α × (OR – 1)/OR, where α is the baseline daily medical costs from 2010 to 2019 per 100 000 person-years. α was calculated among each study group divided by the total person-years for beneficiaries and multiplied by 100 000.
Finally, the results of the sensitivity analyses are presented in Supplementary Table S7 (see online supplementary material for a color version of this table) and we found that our main results are generally consistent with model specifications.
Discussion
This is the first nationwide study to investigate double disparities in temperature-related hospitalization associations among MA beneficiaries in South Korea. This study provides evidence of the disparities in excess risks and costs when considering age, sex, disability status, and causes of hospitalization. MA beneficiaries had a higher heat and cold risk and related excess costs due to extreme temperatures than non-MA beneficiaries, and it was generally more prominent among non-elderly populations (aged <65 years), admissions with mental disorders, males, and people with disabilities.
In general, our findings were consistent with previous studies that showed stronger associations between temperatures and ED hospitalization among MA beneficiaries compared with non-MA beneficiaries. There are several potential explanations for the observed vulnerability of MA beneficiaries. First, fewer people can care for MA beneficiaries than for non-MA beneficiaries (described in more detail in the Supplementary material in ‘1. Korean health security system’). The lack of fundamental social support and assistance increases the risk of adverse health outcomes, making disease detection and management challenging, and exacerbating existing health conditions. Second, low-income individuals are more prone to experiencing poor preexisting health conditions, as they are more likely to face barriers to the adoption of healthy behaviors and management of chronic health conditions due to financial constraints and limited social support networks [29]. Third, they may be unable to afford nutritious food for temperature regulation [30].
Specifically, in terms of risks for emergency hospitalization, considering double disparities among MA beneficiaries by demographic and socioeconomic status, young or middle-aged beneficiaries who are eligible for the MA system showed higher ORs with both heat and cold temperatures. Further, in this group, the risk difference between the MA and non-MA beneficiaries was statistically stronger compared with other age groups (see RORs in Figs 2 and 3). We cautiously conjectured that these results might have been related to occupational and outdoor activities in young and middle-aged populations. Young or middle-aged individuals are more likely to engage in occupational and outdoor activities that could increase the possibility for exposure to extreme temperatures [31–35]. In addition, we carefully surmise that these results could imply that socially marginalized status can play a critical role in contributing to poorer occupational conditions in the young and middle-aged population (e.g. poor working conditions, more frequent outdoor work, occupational stresses from low positions in working places). Lastly, we also would like to suggest that worse preexisting health conditions and a lack of social networks in poor young or middle-aged persons can be additional risk factors [36, 37]. However, these hypotheses should be addressed in further studies with more suitable and extended datasets.
In our study results, the males who were eligible for MA showed a stronger association between temperature exposures and emergency hospitalization than males who were not eligible for MA, and this difference was more prominent for males in cold temperatures than for females. This also could be interpreted partly as occupational differences according to sex among beneficiaries in South Korea, although more in-depth studies are required. In South Korea, economically disadvantaged males tend to work in relatively more intensive industries (such as construction, agriculture, and transportation) that can increase exposure to outdoor extreme temperatures, compared with females and males who are not socially marginalized [6].
In addition, interestingly, MA beneficiaries experienced higher risks of mental disorders than those who were not eligible. We suspect that the compounding effects of poverty-related stressors, such as financial insecurity and social isolation, can increase susceptibility to temperature-related mental health crises [38]. In particular, we observed a positive association between ED admissions for substance abuse disorders and cold temperatures among MA-eligible patients in contrast to the negative association that was observed in the non-eligible group (Supplementary Table S8; see online supplementary material for a color version of this table). According to a previous study, the general population often reduce their consumption of alcohol for pleasure because of the increased financial burden that is associated with heating costs at low temperatures; however, economically marginalized individuals who lack the ability to heat their homes may resort to increased alcohol abuse to heat themselves [39]. Consequently, MA beneficiaries may be more likely to experience alcohol-related disorders, which worsen their overall mental health. This hypothesis should be further investigated in future studies.
MA beneficiaries also showed higher medical costs that were attributable to extreme temperatures than those who were not eligible. With exposure to heat, the attributable costs for most MA subgroups did not show an evident trend. However, for the low temperatures, all subgroups of MA beneficiaries had greater attributable costs than non-MA beneficiaries, especially among elderly people, males, people with disabilities, and people who were diagnosed with cardiovascular and respiratory diseases. In the Korean context, considering that MA beneficiaries are likely to receive medical benefits only when they have severe diseases that lead to ER visits (due to financial barriers that limit regular hospital visits), our estimated attributable costs showed that their admission would require expensive and longer treatments (that might be related to their worse preexisting health status). Therefore, as MA beneficiaries usually suffer from financial burdens, the study showed that they need to be protected from acute health events.
One strength of this study is the use of a large amount of nationwide data that are available for MA beneficiaries and matched controls. Compared with previous studies that had limitations in data and sample size, our study could address these constraints by utilizing large datasets to perform comprehensive research on double disparities. Considering both high and low temperatures, we revealed which subgroups who were eligible for the MA system were vulnerable to temperature exposures by estimating the excess risks and costs that were attributable to temperatures: young and middle-aged populations, males, people with disabilities, and hospitalization with mental disorders.
There are several limitations. First, our study only used data from patients who were admitted to EDs and ignored those who did not visit hospitals, potentially causing underestimation. Second, given the data-memory-capacity constraints of the NHIS analytic center, limitations existed in assuming independence of the risk estimates between the MA and non-MA groups. This assumption could be related to potential biases in the estimations and statistical tests. Finally, this study did not consider various factors that could have potentially influenced the temperature association among MA beneficiaries, due to unavailability in the dataset.
In conclusion, this study revealed double disparities by estimating strong associations between exposure to extreme temperatures and emergency admissions in economically disadvantaged populations. We demonstrated, by using quantitative epidemiological evidence, that the disparities were more evident among MA beneficiaries who were young and middle-aged, males, had disabilities, or faced mental disorder risk factors.
Ethics approval
This study was approved by the Institutional Review Board of Seoul National University, Seoul, South Korea (IRB number E2302/004–002).
Acknowledgements
This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIT) (No. RS-2024–00416848). This work was also supported by the Korea Environment Industry &Technology Institute (KEITI) through the Climate Change R&D Project for the New Climate Regime. This study was also supported by the Korean Ministry of Environment (MOE) (RS-2022-KE002235). We also thank the National Health Insurance Service in South Korea for the data.
Author contributions
A.K. conceptualized the study, performed the statistical analysis, interpreted the results, and wrote the original draft, including visualization. J.P. and C.K. provided the data and contributed to the interpretation of the results and the drafting of the manuscript. H.K. provided knowledge and experience on statistical modeling, assisted with the writing, and provided methodological and empirical insights. W.L. conceptualized and supervised the study, and contributed to the study design, the interpretation of the results, and the drafting of the manuscript. All other authors contributed to the review and revisions of manuscript drafts for important intellectual content.
Use of artificial intelligence (AI) tools
No AI tools were used in collecting or analysing data, producing images, or writing this manuscript.
Supplementary data
Supplementary data are available at IJE online.
Conflict of interest: None declared.
Data availability
The data were analysed under a user agreement with the Korean Health Insurance Service (NHIS) and cannot be made publicly available.
