Nationwide analysis of demographic and socioeconomic inequities in air pollution exposure in Korea, 2006–2019

Abstract

Background

Air pollution is a significant global health challenge, yet few studies have examined how the effects of exposure to environmental pollutants vary across demographic and socioeconomic groups. This study aims to investigate the distribution of air pollutants by demographics and socioeconomic status (SES), accounting for individual monthly residential mobility.

Methods

Individual annual exposures were estimated by linking monthly residential addresses to air pollutant concentrations modeled using the Community Multiscale Air Quality system. Particulate matter (≤ 10 μm, PM₁₀), nitrogen dioxide (NO₂), and ozone (O₃) from 2006 to 2019 were linked to demographic and SES indicators from the Korean National Health Insurance Service for approximately 50 million individuals annually. Exposure disparities across socioeconomic groups were assessed using ANOVA, eta-squared statistics, and generalized estimating equations.

Results

From 2006 to 2019, overall exposures to PM₁₀ and NO₂ declined, whereas O₃ levels increased. PM₁₀ (50.18 µg/m³) and NO₂ (25.01 ppb) concentrations were highest in adults aged 30–34 years, decreasing with age to 47.42 µg/m³ and 22.06 ppb among those aged ≥ 65 years. Conversely, O₃ exposure was lowest in the 25–29-year age group (37.58 ppb) and increased to 39.38 ppb in those aged ≥ 65 years. Eligible individuals (medical aid recipients) had the lowest exposures to PM₁₀ (45.15 µg/m³) and NO₂ (20.12 ppb) but the highest O₃ exposure (40.30 ppb). In the insured populations, self-employed individuals showed higher PM₁₀ and NO₂ exposures compared to employee-insured individuals. Higher SES groups were more likely to be in the highest tertile for PM₁₀ and NO₂ exposure (40.5% of highest-SES self-employed vs. 28.9% of lowest-SES in the high NO₂ tertile), while lower SES groups showed higher O₃ exposure. Spatially, PM₁₀ and NO₂ levels were higher in urban areas, whereas O₃ levels were elevated in rural areas.

Conclusion

SES and demographics intersect to create distinct, pollutant-specific exposure patterns in Korea, contrasting with Western countries where lower-SES groups typically experience higher pollution exposure. By accounting for individual residential mobility, this study provides improved exposure estimates and highlights the importance of considering socioeconomic context in evaluating pollution-related health effects and developing targeted environmental policies.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12889-026-26352-1.

Background

Air pollution is a major global environmental issue due to its adverse health effects [1]. In 2013, the International Agency for Research on Cancer (IARC) of the World Health Organization (WHO) classified air pollution as a Group 1 carcinogen based on evidence of lung cancer caused by particulate matter (PM), a major component of air pollutants [2]. Numerous studies have investigated the health impacts of various air pollutants [3–5]. Jie Chen et al. reported that long-term exposure to PM₂.₅ and PM₁₀ is associated with increased mortality from natural causes, cardiovascular diseases, respiratory diseases, and lung cancer [6]. Similarly, a European cohort study (ELAPSE Project) found that PM₂.₅, NO₂, and black carbon (BC) are strongly positively correlated with natural mortality, as well as deaths related to respiratory and cardiovascular diseases and diabetes and further indicated that air pollutants, including PM₂.₅ and NO₂, are linked to stroke, asthma, chronic obstructive pulmonary disease, gastrointestinal diseases, and lung cancer [7]. More importantly, prior evidence highlights that research on the health effects of air pollution should be further expanded to include socioeconomic impacts [8, 9]. However, few studies have examined how the effects of exposure to environmental pollutants vary across demographic and socioeconomic groups [10, 11].

Moreover, Li et al. emphasized the importance of accurately estimating individual exposure to air pollutants [12]. Individual long-term air pollutant exposure varies with relocation, a factor often neglected in common static monitoring methods [13–15]. The majority of epidemiological research utilizes residential address data from a single time point, typically at enrollment or diagnosis. Yet, this static approach fails to account for dynamic exposure changes resulting from residential mobility, thereby increasing the risk of exposure misclassification. This factor is significant in South Korea, where 3 to 3.7 million individuals (6–7% of the population) relocate every year [16]. In Korea, rapid urbanization has led to distinct residential patterns by socioeconomic status [17], and while PM concentrations have declined due to regulatory efforts, ozone levels have continued to increase [18, 19]. Despite these trends, few studies have comprehensively examined how air pollutant exposure varies across demographic and socioeconomic groups in Korea. This study aims to investigate the distribution of air pollutants by demographics and socioeconomic status (SES), reflecting changes in residential areas. Understanding the intersection between SES and environmental exposure is essential for elucidating the complex epidemiology of environmental health disparities and for developing targeted interventions to mitigate this pressing public health challenge.

Methods

Study population

This nationwide study included nearly the entire Korean population registered in the National Health Insurance Service (NHIS) eligibility database from 2006 to 2019. The NHIS database contains information on date of birth, sex, income level, insurance type, residential address, and related demographic variables. The Korean NHIS operates under the National Health Insurance Act, which mandates universal enrollment and monthly updates. The study aimed to assess average annual exposure to environmental pollutants. Residential addresses were tracked monthly using the NHIS database, and individuals whose addresses were not tracked for 12 consecutive months within a calendar year were excluded. Approximately 1.48–1.85% of individuals were excluded annually due to incomplete residential tracking (with an additional 0.02–0.09% for other reasons) (Table S1, S2). The annual study population comprised approximately 50 million individuals, yielding a total of about 705 million person-records from 2006 to 2019.

Demographic and socioeconomic data in the NHIS database are obtained through linkage with national administrative systems. For employee-insured individuals, employers are legally required to report enrollment and income information, which is cross-verified with National Tax Service records. For self-employed individuals, income levels are verified through National Tax Service records. Residential addresses are updated monthly through linkage with the Resident Registration System operated by the Ministry of Interior and Safety. The accuracy of NHIS data has been validated through comparison with national vital statistics, with concordance correlation coefficients between NHIS mortality data and Statistics Korea vital statistics ranging from 0.997 to 0.999 across municipalities [20].

This study was reviewed and approved for exemption by the Institutional Review Board of Wonju Severance Christian Hospital (CR324323). Given the retrospective design and the use of anonymized data from the NHIS (approval no. REQ202402276-005), the requirement for informed consent was waived. The study was conducted in accordance with the principles outlined in the Declaration of Helsinki.

Assessment of air pollutants

Gridded hourly concentrations of air pollutants across a nationwide domain, simulated at a 9 km horizontal resolution, were generated using a numerical air quality modeling system based on the Community Multiscale Air Quality (CMAQ) model, version 4.7.1 [21, 22]. The modeling system incorporates meteorological conditions, anthropogenic and biogenic emissions, as well as atmospheric chemical and physical processes [21–23](Table 1, Fig. S1, S2). Hourly concentrations of PM₁₀, PM₂.₅, NO₂, SO₂, and CO, as well as the maximum daily 8-hour mean concentration of O₃ [22, 24], were extracted for 250 administrative districts and averaged monthly over the period from January 1, 2006, to December 31, 2019. The monthly average of each air pollutant was matched to the subjects’ addresses to calculate individual monthly exposure levels, which were then averaged to obtain each individual’s annual exposure using administrative district codes (Si-Gun-Gu; N = 250) as the matching identifier. Although all pollutants mentioned above were initially analyzed, this study primarily focused on PM₁₀ and gaseous pollutants (NO₂ and O₃). PM₂.₅ and SO₂ exhibited correlation coefficients of 0.94 and 0.60 with PM₁₀, respectively, while CO showed a correlation coefficient of 0.72 with NO₂ (Table S3). This study presents results for PM₁₀, NO₂, and O₃.

Table 1.

Model performance statistics for the daily mean PM₁₀, NO₂ and O₃ concentrations in South Korea from 2006 to 2019

Air pollutant	Meanobservation	Meanmodel	Meanbias	RMSE	r
PM10(µg/m3)	49.23	49.21	-0.02	2.76	1.00
NO2(ppb)	23.41	23.04	-0.36	0.97	1.00
O3(ppb)	25.33	25.80	0.47	0.89	1.00

SES and demographic indicators

Demographic variables included sex (male, female) and age (categorized into 5-year intervals starting from 0 to 4 years; however, the upper truncation varied by analysis type), while SES was defined based on NHIS eligibility and health insurance premiums. Subscription for national health insurance in Korea is categorized into eligible individuals (those qualified for medical aid under the national welfare system), employee-insured, and self-employed insured [25]. Eligible individuals receive medical services through Korea’s national welfare system and do not pay health insurance premiums. Employee-insured individuals include employees, employers, and their dependents, whereas self-employed insured individuals comprise small business owners, freelancers, the unemployed, and their dependents. Health insurance premiums are calculated based on salaries for employees and on income and assets for self-employed individuals. In this study, socioeconomic status (SES) was categorized using two distinct variables: a detailed 11-level variable (comprising Medical Aid beneficiaries, five premium-based quintiles of the employee-insured, and five premium-based quintiles of the self-employed) and a simplified binary variable (employee-insured vs. self-employed). Geographic variables were defined at two levels: (1) Region, consisting of 250 administrative districts used for air pollution linkage; and (2) Urbanization, a 3-level categorical variable (urban, semi-urban, rural) derived from these districts based on Korea’s administrative divisions as a proxy for urbanization. Age, SES, and residence were allocated as of December of each calendar year.

Statistical analysis

Descriptive statistics were used to describe the general characteristics of the study population at the baseline year of 2006. The proportion of study participants by SES across age groups and the age composition within each SES category were assessed to characterize the distribution of SES levels within each age group in Korea.

Pollutant concentrations were categorized into annual tertiles based on the study population’s annual distribution to account for temporal changes in population size and exposure variability in each year from 2006 to 2019. The resulting annual cumulative exposure tertile thresholds were presented as minimum, 33rd percentile (1st tertile cutoff), 67th percentile (2nd tertile cutoff), and maximum values. Mean pollutant exposure levels were calculated by sex, age group, and SES was stratified by eligibility status (medical aid recipients), self-employed, and employee-insured populations, with quintiles used for the self-employed and employee-insured groups to assess exposure differences by income level. Trends in air pollutant concentrations from 2006 to 2019 were assessed to examine differences between mobility-adjusted individual exposure levels and ambient air pollution concentrations over time. Exposure distributions of PM₁₀, NO₂, and O₃ were assessed for medical aid recipients and across five SES quintiles for self-employed and employee-insured populations. Within each SES group, participants were assigned to annual exposure tertiles based on year-specific concentration distributions, and the proportion of individuals in each tertile (low, medium, high) was calculated.

To assess the influence of residential area characteristics, we analyzed cumulative mean pollutant exposure levels across three urbanization categories based on annual administrative status: ‘Gu’ (Urban; districts), ‘Si’ (Semi-urban; cities), and ‘Gun’ (Rural; counties). This classification approach captured administrative changes. Average concentrations of PM₁₀, NO₂, and O₃ were calculated for each group, with geographic variations visualized using choropleth maps.

To assess the independent effects of demographic and socioeconomic factors on air pollutant exposure, we conducted a series of one-way analyses of variance (ANOVA) and t-test separately for each factor. To evaluate the practical significance of the observed differences beyond statistical significance (which is sensitive to large sample sizes), we calculated the effect size using eta-squared (η²). Effect sizes were categorized as negligible (η² < 0.01), small (0.01 ≤ η² < 0.06), medium (0.06 ≤ η² < 0.14), or large (η² ≥ 0.14) [25, 26].

To account for temporal autocorrelation in annual repeated measurements, we applied population-weighted generalized estimating equations (GEE) assuming a Gaussian distribution with an AR(1) correlation structure. Due to computational constraints associated with approximately 700 million person-years of individual-level data, GEE models were fitted on aggregated data stratified by region (250), sex, age group (up to ≥ 85), and SES. Weighting by person-years was applied to account for the varying population sizes across strata. The main models included urbanization level (3), sex, age group (up to ≥ 85), SES (11 categories), and calendar year as covariates, with PM₁₀, NO₂, and O₃ concentrations as dependent variables. Type III Wald Chi-Square tests were used to assess the overall statistical significance of categorical variables. Additionally, to examine effect modification by urbanization, we conducted stratified analyses comparing self-employed versus employee-insured individuals (grouping income quintiles) within each urbanization level (urban, semi-urban, and rural).

All statistical analyses were performed using SAS (version 9.4). Statistical significance was set at p < 0.001 and/or effect size η² unless otherwise specified.

Results

Socio-demographic characteristics

The study population was evenly distributed by sex. Employee-insured individuals constituted the largest group, followed by the self-employed and medical aid recipients (Table 2). Across age groups, SES distribution varied by insurance type (Table 3). Among the employee-insured, the predominant income quintile varied non-linearly with age: middle-to-high quintiles (3rd–5th) predominated among those aged 0–19 years, lower quintiles (1st–2nd) among those aged 20–29 years, higher quintiles (3rd–5th) among those aged 30–49 years, and lower-to-middle quintiles among those aged ≥ 50 years. Among the self-employed, higher quintiles (4th–5th) predominated from age 45 onward. Individuals aged ≥ 65 years had the highest proportion of medical aid recipients (10.25%), whereas those aged 25–29 years had the lowest (0.82%). Among the self-employed insured, the 40–44-year group predominated in the lower SES quintiles.

Table 2.

Distribution of populations by general characteristics in 2006

Variables		N	%	Subtotal (%)
Sex	Male	24,321,238	50.12
	Female	24,208,036	49.88
Age group(years)	0–4	1,929,135	4.22
	5–9	3,126,844	6.84
	10–14	3,538,912	7.74
	15–19	3,198,929	6.99
	20–24	3,502,062	7.66
	25–29	4,005,925	8.76
	30–34	4,195,908	9.17
	35–39	4,559,491	9.97
	40–44	4,163,920	9.10
	45–49	4,260,518	9.31
	50–54	3,202,756	7.00
	55–59	2,343,831	5.12
	60–64	1,957,415	4.28
	65+	1,759,801	3.85
Socioeconomic status	Eligible (medical aid)	1,782,921	3.67	3.67
	1st Self-employed	3,550,348	7.32	37.35
	2nd Self-employed	3,608,570	7.44
	3rd Self-employed	3,632,622	7.49
	4th Self-employed	3,656,754	7.54
	5th Self-employed	3,677,464	7.58
	1st Employee	5,515,189	11.36	56.55
	2nd Employee	5,420,198	11.17
	3rd Employee	5,462,070	11.26
	4th Employee	5,509,418	11.35
	5th Employee	5,536,540	11.41
	Unknown	1,177,180	2.43	2.43
Urbanization	Urban	29,750,455	61.30
	Semi-urban	14,059,299	28.97
	Rural	4,719,427	9.72

*Data are shown as person number (%)

Table 3.

Proportion of participants’ socioeconomic status by age group in 2006

	Age group (years)	Socioeconomic status											Total
		Eligible	1st SE	2nd SE	3rd SE	4th SE	5th SE	1st E	2nd E	3rd E	4th E	5th E
Proportion of SES within each age group	0–4	3.27	6.57	8.39	7.47	5.89	3.68	10.78	13.70	17.37	14.56	8.31	100
	5–9	4.84	7.07	8.82	8.41	7.07	4.70	10.94	9.95	13.24	14.75	10.22	100
	10–14	5.27	7.82	9.16	9.04	8.47	6.18	11.92	7.93	9.22	12.80	12.18	100
	15–19	5.45	8.18	8.48	8.83	9.44	7.98	12.84	7.80	7.13	9.86	14.01	100
	20–24	1.96	6.77	7.34	7.33	8.27	8.12	15.48	14.43	10.74	8.19	11.38	100
	25–29	0.82	6.82	5.88	5.49	5.29	5.48	12.68	17.80	16.64	12.76	10.34	100
	30–34	1.12	7.44	7.85	7.21	6.02	5.03	9.71	14.13	16.29	14.13	11.06	100
	35–39	1.97	8.52	9.31	9.01	7.76	6.25	9.29	10.44	12.28	13.49	11.69	100
	40–44	2.96	9.23	9.74	9.79	9.38	8.13	9.75	8.63	8.51	11.17	12.70	100
	45–49	3.34	8.14	8.86	9.35	10.27	10.58	11.64	9.12	7.19	8.47	13.04	100
	50–54	3.29	6.68	6.52	7.44	9.16	12.47	13.74	12.16	8.72	7.06	12.76	100
	55–59	3.53	5.50	5.02	5.87	7.69	12.57	14.80	15.04	12.22	8.09	9.66	100
	60–64	4.56	5.67	4.55	5.34	6.82	11.57	13.87	14.17	14.21	11.39	7.85	100
	65Over	10.25	7.84	4.85	5.20	5.84	7.97	9.43	8.85	11.25	14.68	13.87	100
Proportion of age group within each SES	0–4	3.43	3.46	4.35	3.85	3.01	1.87	3.66	4.73	5.95	4.95	2.81
	5–9	8.28	6.08	7.46	7.06	5.90	3.90	6.05	5.60	7.40	8.17	5.63
	10–14	10.26	7.65	8.81	8.63	8.04	5.83	7.50	5.08	5.86	8.07	7.64
	15–19	9.59	7.23	7.37	7.63	8.11	6.81	7.31	4.51	4.10	5.62	7.94
	20–24	3.72	6.45	6.87	6.83	7.65	7.46	9.49	9.01	6.65	5.03	6.95
	25–29	1.78	7.42	6.30	5.84	5.59	5.76	8.88	12.69	11.78	8.95	7.22
	30–34	2.55	8.49	8.82	8.05	6.68	5.55	7.14	10.57	12.09	10.40	8.10
	35–39	4.91	10.66	11.46	11.02	9.43	7.55	7.48	8.56	9.99	10.88	9.38
	40–44	6.77	10.60	11.01	10.99	10.47	9.02	7.21	6.49	6.35	8.27	9.36
	45–49	7.84	9.58	10.26	10.74	11.73	12.01	8.82	7.03	5.49	6.42	9.83
	50–54	5.79	5.90	5.67	6.43	7.85	10.64	7.82	7.04	5.01	4.02	7.23
	55–59	4.56	3.57	3.20	3.72	4.85	7.87	6.18	6.39	5.15	3.38	4.02
	60–64	4.92	3.07	2.43	2.83	3.59	6.06	4.84	5.03	5.01	3.98	2.73
	65Over	25.60	9.83	5.98	6.37	7.11	9.65	7.62	7.27	9.17	11.87	11.16
	Total	100	100	100	100	100	100	100	100	100	100	100

SE self-employed, E employee, SES socioeconomic status, eligible refers to recipients of medical aid

Annual distribution of air pollutants

Cumulative exposure to air pollutants, accounting for residence mobility, showed consistent trends from 2006 to 2019. PM₁₀ and NO₂ exposures declined over time, whereas O₃ exposure increased. Over the 14-year study period, PM₁₀ and NO₂ concentrations showed a consistent downward trend, decreasing by approximately 29% and 15% respectively, whereas O₃ levels exhibited a substantial upward trajectory (Fig. 1). Individual exposures to PM₁₀ and NO₂ exceeded ambient air concentrations, whereas O₃ levels were lower than ambient levels (Fig. 1A). The distribution of annual exposure—including minimum, tertiles (33rd and 67th percentiles of the annual study population), and maximum values—showed consistent decreases in PM₁₀, stable or slightly lower NO₂ levels, and increases in O₃ over time (Fig. 1B). PM₂.₅, SO₂, and CO exhibited trends similar to PM₁₀ and NO₂, whereas O₃ showed an opposing pattern over the study period (Supplementary Fig. S3).

Fig. 1.

Trends in air pollutant concentrations (2006–2019). (A) Individual and atmospheric trends. Individual exposure levels account for residential mobility, whereas atmospheric concentrations represent regional ambient air levels. The same distinction applies to PM₁₀, NO₂, and O₃. η² values indicate the variance explained by calendar year. (B) Annual cumulative exposure tertile thresholds determined by the study population distribution

Exposure to air pollutants by demographics and SES

Exposure differences by sex were negligible for all pollutants, although statistically significant due to the large sample size (Fig. 2A). Exposure levels varied by age: PM₁₀ and NO₂ concentrations peaked in the 30–34 age group and progressively decreased with age, whereas O₃ exposure showed an inverse pattern, increasing with age (Fig. 2B). In contrast, O₃ exposure was lowest in the 25–29 age group (37.58 ppb) and increased with age, reaching 39.38 ppb among those aged 65 and older (Fig. 2B). Similar age-related trends were observed across the same SES groups (Supplementary Fig. S4). There were notable differences in air pollutant exposure levels across SES and insurance types. Eligible individuals experienced the lowest exposures to PM₁₀ (45.15 µg/m³) and NO₂ (20.12 ppb), but the highest exposure to O₃ (40.30 ppb). Among both self-employed and employee-insured groups, PM₁₀ and NO₂ exposures rose with higher SES, whereas O₃ exposure decreased. The self-employed group consistently exhibited higher exposures to PM₁₀ and NO₂, and lower exposures to O₃, compared with the employee-insured group across all income levels (Fig. 2C). PM₁₀ and NO₂ concentrations followed a clear urbanization gradient, highest in urban areas and lowest in rural areas. Conversely, O₃ concentrations exhibited the opposite pattern, peaking in rural areas (Fig. 3).

Fig. 2.

Air pollutant concentrations by demographics and SES (2006–2019). (A) By sex, (B) By age group, (C) By SES. η² values are provided for each factor. Eligible refers to individuals receiving medical aid through Korea’s national welfare system. SES, socioeconomic status; SE, self-employed; E, employee. Numbers 1–5 indicate income quintiles (1 = lowest, 5 = highest)

Fig. 3.

Spatial distribution of air pollutants across 250 municipalities by urbanization level in South Korea. Maps display PM₁₀ (µg/m³), NO₂ (ppb), and O₃ (ppb) concentrations stratified by urbanization classification (urban, semi-urban, and rural; Gu/Si/Gun). Numbers indicate mean concentrations for each urbanization category, with national-level distribution shown in the final column. η² values represent the variance explained by urbanization level and administrative region

The distribution of exposure tertiles across SES groups further demonstrated these disparities. Eligible individuals and lower SES quintiles were disproportionately represented in the lowest tertile for PM₁₀ and NO₂, but in the highest tertile for O₃. Conversely, the highest SES quintiles (5th) within both insurance groups had the greatest proportions in the highest PM₁₀ and NO₂ tertile and the lowest O₃ tertile (Fig. 4). These SES-related exposure patterns were consistent across age groups and urbanization levels (Fig. S4-6).

Fig. 4.

Proportion of exposure to air pollutants by SES (2006–2019). Stacked bars show the proportion of exposure tertiles (1st, 2nd, 3rd) within each SES group. Tertiles were determined annually based on year-specific exposure distributions of the study population. Eligible refers to individuals receiving medical aid through Korea’s national welfare system. SES, socioeconomic status; SE, self-employed; E, employee

In terms of the magnitude of variance explained, calendar year showed a large effect size for PM₁₀ and O₃, and a medium effect size for NO₂ (Fig. 1A). Administrative region (250) demonstrated a large effect size across all three pollutants, and urbanization level (3) also exhibited a large effect size for gaseous pollutants (NO₂, O₃) but a small effect size for PM₁₀ (Fig. 3). Conversely, sociodemographic factors had limited influence on exposure variation. Sex showed a negligible effect for all pollutants (Fig. 2A), while age group showed small effects (Fig. 2B). Socioeconomic status (11) showed negligible effects for PM₁₀ and O₃, with a small effect observed for NO₂ (Fig. 2C).

The GEE analysis, accounting for temporal autocorrelation, confirmed that SES remained significantly associated with all three pollutants after adjusting for age, sex, year, and urbanization level (P < 0.001). In contrast, sex was not significantly associated with any pollutant concentrations (P > 0.05) (Table S4). When stratified by urbanization level the exposure disparity between self-employed and employee-insured individuals varied by urbanization level. In urban areas, the self-employed had significantly higher NO₂ exposure than employees (β = 2.79), whereas in rural areas, they had lower exposure (β = -0.58) (Table S5).

Discussion

In this study, we assessed the distribution of air pollutant levels in Korea over a 14-year period (2006–2019), accounting for individual-level residential mobility monthly throughout the entire study period. As shown in Fig. 1A, PM₁₀ and NO₂ concentrations showed a downward trend, whereas O₃ exhibited an upward trajectory over the study period. While the trends were consistent, the absolute levels of individual exposure (mobility-adjusted) differed from atmospheric measurements. Specifically, individual exposure to PM₁₀ and NO₂ was higher, reflecting population density in urban areas. Notably, even though we accounted for the substantial residential mobility of approximately 6–7% of the population annually [16], the temporal patterns of individual exposure closely mirrored the atmospheric trends. This indicates that while residential mobility contributes to the diversity of individual exposure estimates, it did not alter the overall longitudinal trends of population-wide exposure. Overall, exposures to PM₁₀ and NO₂ decreased over the study period, whereas O₃ levels increased. PM₁₀ and NO₂ exposure levels were higher among adults aged 25–34 years and in urban areas, whereas O₃ exposure increased with age and was higher in rural areas. Eligible individuals had the lowest PM₁₀ and NO₂ exposures but the highest O₃ levels compared with other socioeconomic groups.

The SES quintiles representing the largest proportion of employee-insured individuals gradually increased with age, such that the first income quintile was the most prevalent among those in their 50s. This pattern may reflect Korea’s employment structure, in which many people in their 50s retire from previous jobs and begin new careers, leading to a decline in income [26]. For both employee-insured and self-employed–insured individuals, the income distribution among those aged 0–19 was similar to that of those aged 30–49. This similarity is likely because individuals aged 0–19 are typically dependents of those aged 30–49, consistent with the fact that the mean maternal age at first childbirth was 29 years in 2006 [27]. Among both employee-insured and self-employed–insured individuals in older age groups, higher SES quintiles were more common, likely reflecting sustained income gains and greater asset accumulation following reemployment in their 50s [27, 28].

A global study of more than 50,000 individuals across 13,160 cities reported that from 2000 to 2019, PM₂.₅ exposure decreased by an average of 0.2% per year, while O₃ exposure increased by an average of 0.6% per year [29]. In Korea, our 14-year analysis from 2006 to 2019 showed more marked variations, as PM₂.₅ and PM₁₀ decreased by an average of 1.4% and 2.3% per year, respectively, while O₃ increased by an average of 1.6% per year. Globally, NO₂ exposure increased by an average of 0.4% per year [29], whereas in Korea, it decreased by an average of 1.2% per year during the study period. The global increase in NO₂ is primarily driven by economic growth and industrialization in regions such as sub-Saharan Africa, the Middle East, and South Asia, whereas some North American and European countries exhibited decreasing trends similar to those observed in Korea [29]. In Korea, the substantial reductions in PM₁₀ and NO₂ can be attributed to comprehensive air quality policies implemented since the early 2000s, Key measures included the Special Act on Metropolitan Air Quality Improvement (2005), nationwide adoption of stricter vehicle emission standards, expansion of compressed natural gas (CNG) buses, and enhanced industrial emission controls [18, 30].

The concurrent increase in O₃ reflects the complex, nonlinear photochemistry of tropospheric ozone formation. As NOx levels decline, ozone is less efficiently scavenged by nitric oxide—a phenomenon known as the NOx titration effect—resulting in net increases in ambient O₃ concentrations [31, 32]. In contrast, rural areas with limited transportation infrastructure and lower population density exhibit higher O₃ concentrations due to this photochemical mechanism. This inverse relationship is evident in our data, where NO₂ and O₃ exhibited a strong negative correlation (r = − 0.76; Supplementary Table S3). Additionally, climate-related factors such as rising temperatures and stronger solar radiation further enhance photochemical ozone production [33, 34]. This inverse pattern of increasing O₃, despite reductions in precursor pollutants, underscores the challenges of air quality management and highlights the importance of integrated strategies targeting multiple pollutants simultaneously.

While initial unadjusted analyses (ANOVA) suggested statistically significant gender differences, our comprehensive GEE analysis revealed that gender was not an independent predictor of pollutant exposure after adjusting for age, SES, and urbanization level (P > 0.05). This discrepancy suggests that the apparent gender differences observed in descriptive statistics were likely confounded by other sociodemographic factors, indicating that gender itself contributes minimal variance to residential exposure levels in Korea. By age group, individuals aged 30–34 years had the highest exposure to PM₁₀ and NO₂, with levels decreasing progressively with age, while those aged 25–29 years had the lowest exposure to O₃, which increased with advancing age. These patterns likely reflect Korea’s demographic characteristics, where younger populations are more concentrated in urban areas, whereas older populations are more likely to reside in rural regions [35–37]. With respect to SES, exposure to PM₁₀ and NO₂ was greater among higher-SES groups, whereas O₃ exposure increased as SES decreased. Notable differences in pollutant exposure were observed across SES, a pattern that may be explained by the tendency of high-income populations to live in urban centers with dense traffic and elevated PM concentrations, while lower-income populations are more likely to reside in suburban or rural areas where O₃ levels are relatively higher [17, 38].

PM₁₀ and NO₂ concentrations were highest in high-density urban areas, followed by semi-urban areas, and lowest in rural areas, primarily due to variations in traffic density and industrial activity [39]. Conversely, O₃ exposure was highest in rural areas, followed by semi-urban areas, and lowest in urban areas [40, 41]. Previous studies in Western contexts, particularly North America and Europe, have generally reported that socioeconomically vulnerable (lower-SES) groups experience higher exposure to air pollutants due to residential proximity to industrial facilities and major roadways [42–44]. Our findings in Korea reveal a contrasting trend. Similar to recent observations in other rapidly urbanizing regions, we found that higher-SES populations are more exposed to ambient NO₂ and PM₂.₅. This pattern is attributed to rapid urbanization and the concentration of economic activity in polluted urban centers [45]. This discrepancy may reflect urban development trends in Korea, where areas with higher SES have greater access to public transportation and road infrastructure, which in turn increases traffic volumes and associated pollution levels [46, 47].

The results indicate that areas with high PM₁₀ and NO₂ levels are more densely populated, have predominantly younger residents and higher SES levels, whereas cities with elevated O₃ levels tend to be smaller, have older populations, and lower SES. This pattern in Korea can be attributed to the nation’s distinctive urban development trajectory during rapid industrialization in the 1980s, when major roads and residential areas were often constructed in proximity [48, 49]. During this period, proximity to main roads was highly valued for commuting convenience, driving up housing demand and prices in these areas [49]. As a result, higher-SES populations tended to settle near major transportation corridors and industrial centers, leading to increased exposure to traffic-related pollutants such as PM₁₀ and NO₂. In contrast, rural areas with limited transportation infrastructure and lower population density exhibit higher O₃ concentrations due to reduced NO_x titration and conditions favorable for photochemical ozone formation [31, 32]. Additionally, biogenic volatile organic compounds emitted from vegetation in rural and forested areas react with NOx under sunlight, further enhancing photochemical O₃ production [50]. These spatial patterns of pollutant distribution reflect not only geographic and environmental factors but also socioeconomic characteristics of residents.

Specifically, we observed a distinct divergence in exposure disparities by urbanization level. In urban areas, self-employed individuals had significantly higher NO₂ exposure compared to employee-insured workers. This pattern aligns with socio-spatial segregation characteristics in Korean cities. Self-employed households are more likely to reside in mixed-use buildings within commercially dense districts or old city centers characterized by high traffic volume [51], whereas employee-insured households tend to cluster in residential-focused districts or new towns designed with greater setbacks from pollution sources [49]. Conversely, in rural areas, self-employed individuals showed lower NO₂ exposure than employees. This reversal relates to the spatial structure of rural communities, where administrative town centers concentrate essential services, local traffic, and the residences of office workers (e.g., civil servants, teachers) [52]. In contrast, the self-employed population, largely comprising farmers, is widely distributed across less populated outskirts with minimal traffic emissions.

Our findings suggest that differences in pollutant exposure between self-employed and employee-insured populations should be considered in the development of environmental health policies. Understanding these Korea-specific patterns is critical for designing targeted interventions. Policies aimed at reducing traffic-related pollutants in affluent urban areas could protect large segments of the workforce, while O₃ mitigation efforts should focus on rural regions with older, lower-SES populations. Future investigations should examine whether these socioeconomic–environmental dynamics are present in other rapidly industrializing Asian countries and how ongoing urbanization and infrastructure changes may alter exposure distributions.

Importantly, policies must extend beyond outdoor air quality to consider ‘effective exposure’. While our study relied on outdoor estimates, medical aid recipients may face compounded risks due to housing conditions. Previous studies suggest that older or lower-quality housing, often inhabited by lower-SES groups, tends to have poorer airtightness and lacks air conditioning, leading to higher infiltration rates of outdoor ozone into indoor environments [53]. Therefore, despite the relatively lower outdoor O₃ concentrations in some urban low-SES areas compared to rural areas, the actual physiological exposure for these residents may be exacerbated by poor housing defenses.

This study provides a comprehensive analysis of air pollutant exposure levels across demographics and SES in Korea from 2006 to 2019, incorporating monthly residential mobility. This is among the few nationwide studies to utilize individual-level monthly address updates over such extended observation periods, substantially reducing exposure misclassification and providing more accurate estimates of long-term disparities in pollutant exposure across sociodemographic groups. Furthermore, the use of the CMAQ model enabled high-resolution spatial assessment of multiple pollutants, including PM₁₀, NO₂, and O₃, across urban, semi-urban, and rural areas.

However, several limitations should be noted. First, exposure estimates relied on residential location without daily time-activity patterns, occupational exposure, or workplace location [54, 55]. Workplace addresses were not consistently available in the NHIS database, limiting our ability to assess exposure at non-residential workplace or explore these factors. This may have caused exposure misclassification. Second, indoor pollutant levels and housing conditions were not assessed [56, 57]. Third, the study did not account for lifestyle-related confounders, including tobacco use and secondhand smoke exposure, which influence both indoor and outdoor pollutant levels [58]. These unmeasured factors may have biased our exposure-outcome associations. Future research should incorporate personal time-activity surveys, occupational exposure data, and housing characteristics to better characterize individual-level exposure variability. Despite these limitations, incorporating monthly address changes substantially improved upon conventional approaches using fixed baseline addresses, reducing residential mobility bias in this study. This study sheds light on distinct sociodemographic inequalities in environmental pollution exposure in Korea, providing a strong foundation for evidence-based policies aimed at protecting vulnerable populations and informing future environmental health research in rapidly developing nations.

Conclusions

This long-term nationwide study, tracking monthly residential mobility for nearly the entire Korean population over 14 years, demonstrates pollutant-specific environmental inequalities shaped by age, SES, and urbanization in Korea. PM₁₀ and NO₂ exposure levels decreased over time, whereas O₃ levels increased. Younger, higher-SES, urban populations experienced greater exposure to PM₁₀ and NO₂, whereas older, lower-SES, rural populations were exposed to higher levels of O₃. Policies and interventions should account for these disparities when evaluating the adverse health effects of air pollution and protecting vulnerable populations. Future studies are warranted to refine exposure assessments by incorporating indoor air quality, occupational exposure, and additional pollutants as well as examine health outcomes associated with these exposure inequalities.

Authors’ contributions

SY Kim: Conceptualization, Formal analysis, Investigation, Software, Data Curation, Writing–Original Draft, Writing–Review & Editing, Visualization. SJ Nam Goung: Conceptualization, Writing – Original Draft. YH Kang, S Kim, SY Heo: Resources, Investigation. MJ Koo: Validation, Resources, Software. JY Kim: Conceptualization, Methodology, Software, Validation, Supervision. KS Jeong: Conceptualization, Methodology, Funding acquisition, Software, Writing – Review & Editing, Project Administration, Supervision. All authors have read and agreed to the published final version of the manuscript.

Funding

This study was supported by the Environmental Health Center operating budget (Health Information Big Data Environmental Health Center) (grant no. 2024-71-0171). Additional support was provided by the “Core Technology Development Project for the Prevention and Management of Environmental Diseases”, administered by the Korea Environmental Industry & Technology Institute (KEITI) through the Yonsei University Wonju Industry–Academic Cooperation Foundation (grant no. 2025-71-0051).

Data availability

The data supporting this study’s findings were obtained from Korea’s NHIS. Restrictions apply to the availability of these data, which were used under license for the current study. The data are not publicly available due to privacy and ethical restrictions imposed by the NHIS data use agreement. Access to the data is available only from the NHIS.

Declarations

Ethics approval and consent to participate

This study was reviewed and approved for exemption by the Institutional Review Board of Wonju Severance Christian Hospital (CR324323). Given the retrospective design and the use of anonymized data from the NHIS (approval no. REQ202402276-005), the requirement for informed consent was waived. The study adhered to the principles outlined in the Declaration of Helsinki.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.