Association between ambient air pollution and prevalence of myopia in Korean men

Abstract

Background

Ambient air pollution may contribute to the growing burden of myopia, but evidence from large population-based studies remains limited. We evaluated the association between long-term exposure to ambient air pollution and the prevalence and severity of myopia in young Korean men.

Methods

We analyzed nationwide military conscription examination data from 2010 to 2020, including 1,671,826 nineteen-year-old men residing in seven metropolitan cities. Participants without educational attainment data or living outside major urban areas were excluded. We examined the association between district-level 3-year average concentrations of particulate matter (PM₂.₅ and PM₁₀) and nitrogen dioxide (NO₂) with the prevalence and severity of myopia using logistic regression models adjusted for individual and district-level covariates. Sensitivity analyses considered 1- and 6-year exposure averages.

Results

Higher NO₂ exposure was associated with increased odds of myopia (OR per IQR: 1.262; 95% CI: 1.254–1.270), with stronger associations for high myopia (OR per IQR: 1.354; 95% CI: 1.342–1.366), in models adjusted for individual-level covariates. A clear dose-response pattern was observed across NO₂ quartiles. PM₁₀ also showed positive but weaker and less consistent associations, with slightly stronger links to low myopia. In contrast, PM₂.₅ demonstrated consistent inverse associations, likely reflecting temporal confounding due to limited data availability. Sensitivity analyses yielded similar results.

Conclusions

Long-term exposure to NO₂ and PM₁₀ may contribute to both the development and progression of myopia, with NO₂ showing particularly strong associations for high myopia. Further longitudinal studies are warranted to clarify these relationships and inform preventive strategies.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12889-026-26599-8.

Background

Myopia, commonly known as nearsightedness, is a visual abnormality in which the resting eye focuses images of distant objects in front of the retina, resulting in blurred vision. It has emerged as a major global public health concern, with prevalence rising markedly in recent decades. It is estimated that by 2050, approximately 50% of the global population will be affected by myopia [1].​ East Asian countries report some of the world’s highest myopia burdens among young people. In South Korea, recent surveys show that 70–80% of adolescents have myopia, and 15–20% have high myopia [2].​ Individuals with myopia face elevated risks of sight-threatening pathologic complications such as retinal detachment, myopic maculopathy, glaucoma, and cataracts [3, 4] 2020; Du et al., 2024). Although genetic factors are the most well-recognized contributors to myopia [5–7], the rapid increase in its prevalence cannot be solely explained by genetics. Mounting evidence argues against the purely genetic etiology of the modern myopia epidemic [8], and underscores the need to identify modifiable environmental contributors. Traditional environmental risk factors for myopia include intensive near work, such as prolonged reading or screen use [9, 10] and insufficient outdoor activity during childhood [8]. Studies have shown that greater time spent outdoors exerts a protective effect, presumably via increased exposure to bright light, which stimulates dopamine release in the retina and slows axial elongation of the eye [11]. Additionally, several epidemiological studies have reported a higher incidence of myopia in individuals with autoimmune diseases such as Kawasaki disease, type 1 diabetes mellitus, systemic lupus erythematosus, and uveitis [12, 13], as well as associations with allergic conditions, including allergic rhinitis, conjunctivitis, and atopic dermatitis [14, 15]. These findings suggest a potential role of inflammation in the pathogenesis of myopia.

Ambient air pollution, an established pro-inflammatory environmental factor, has recently attracted attention as a potential contributor to myopia [16]. Proposed mechanisms include pollutant-induced ocular surface inflammation (e.g., allergic conjunctivitis) that alters corneal curvature, oxidative stress that disrupts scleral remodeling, and behavioral reduction in outdoor time owing to smog [16]. Experimental evidence also supports a causal relationship between particulate matter (PM) and myopia. In an animal study, Park et al. [17] demonstrated that PM exposure impaired mitochondrial function in corneal epithelial cells, increased oxidative stress, and triggered pro-inflammatory cytokine production, ultimately leading to cell death. These inflammatory responses disrupt ocular homeostasis and promote axial elongation of the eye. In a complementary animal study, Wei et al. [18] reported that 3-week-old hamsters exposed to high levels of PM_2.5 developed significantly greater myopic refraction than unexposed controls. Collectively, these findings support the hypothesis that PM exposure influences ocular development through inflammatory mechanisms, thereby contributing to the onset or progression of myopia.

In addition to animal studies, several epidemiological investigations have suggested that chronic exposure to air pollution is associated with an increased risk of myopia in children. A cohort study utilizing the Taiwanese National Health Insurance Database showed that districts with higher PM_2.5 and nitrogen dioxide (NO₂) concentrations experienced increased myopia incidence among 97,306 children aged 6–12 over a 12-year period [18]. In Spain, Dadvand et al. [19] observed that children aged 7–10 exposed to elevated traffic-related air pollution (measured by PM_2.5 absorbance) had increased odds of requiring spectacles for myopia. More recently, a nationwide cross-sectional study in China reported positive correlations between ambient pollutants (PM₁₀, PM_2.5, and NO₂) and myopic refractive errors in school-aged populations [20]. Despite these findings, epidemiological studies examining air pollution and myopia have several limitations. First, most studies were conducted with relatively small samples, typically ranging from a few thousand to fewer than one hundred thousand participants, which may limit the generalizability of their findings. Second, the investigations primarily focused on school-aged children whose refractive status remains in flux due to ongoing ocular development. Longitudinal research suggests that myopia generally stabilizes between the ages of 18 and 21 [21–23]. Therefore, pediatric cohorts may not capture the final refractive outcomes or long-term myopia burden. Third, previous studies have often relied on indirect or incomplete assessments of myopia. Specifically, some studies used health insurance records [18], while others used spectacle prescription as proxy indicators [19], and some relied solely on visual acuity tests without incorporating objective refraction [20]. Consequently, the accuracy and objectivity in determining true refractive status, particularly myopia, may be limited.

To address these limitations, this study leveraged a nationwide dataset derived from South Korean military conscription examinations conducted between 2010 and 2020. This dataset encompasses nearly the entire population of 19-year-old male citizens, offering an exceptionally large, uniform, and demographically representative sample. Crucially, it includes standardized refractive error measurements obtained at an identical age (late adolescence) for each participant. This study aimed to examine the associations between long-term exposure to ambient PM and NO₂ and myopia prevalence and severity in late adolescence by using a large, nationally representative dataset.

Methods

Data source

This study utilized nationwide data from military conscription examinations conducted in South Korea. In accordance with the Military Service Act, all male citizens are subject to conscription and are required to undergo a physical examination during the year they turn 19. These examinations are not routine health screenings but are specifically designed to assess physical and functional impairments to determine eligibility for military service. The dataset includes comprehensive demographic and health-related variables, such as district-level residential addresses, physical measurements (height, weight, body mass index [BMI], and both systolic and diastolic blood pressure), results from visual assessments (including visual acuity measured using Snellen charts and spherical and cylindrical power by a non-cycloplegic autorefraction device), and educational attainment (final education level recorded as of 2021, matched automatically from the national educational database using candidates’ social security numbers).

Study population and area

We obtained data on 3,712,827 19-year-old males who underwent mandatory physical examinations for military service in South Korea between 2010 and 2020. Among these, 44 individuals without educational attainment data were excluded, resulting in a total of 3,712,783 participants. To enhance spatial validity and minimize exposure misclassification, we restricted the final study population to individuals residing in South Korea’s seven metropolitan cities, where population density is consistently high and intra-city environmental conditions are relatively homogeneous. The final sample comprised 1,671,826 individuals residing in Seoul, Busan, Daegu, Incheon, Gwangju, Daejeon, and Ulsan (Supplementary Fig. S1). The geographical distribution of these cities is shown in Supplementary Fig. S2.

Definition of myopia

The spherical equivalent (SE) was calculated using spherical and cylindrical refractive values according to the following formula [24]:

Myopia was defined as a SE of ≤ − 0.50 diopters (D) in at least one eye [25]. Although definitions of myopia subtypes vary slightly across studies, we adopted the classification proposed by the International Myopia Institute. Accordingly, high myopia was defined as an SE of ≤ − 6.00 D in one or both eyes, while low myopia was defined as an SE of ≤ − 0.50 D and > − 6.00 D [25]. We also considered two additional outcome variables. First, myopia was treated as a binary outcome using the standard threshold of SE of ≤ − 0.5 D. Second, among individuals with myopia, we further categorized severity into low myopia (SE − 0.5 D to − 6.0 D) and high myopia (SE ≤ − 6.0 D), and analyzed this as an ordinal outcome variable.

Exposure assessment

We estimated long-term individual-level concentrations of PM₁₀, PM_2.5, and NO₂ using a previously validated air pollution prediction model [26, 27]. This nationwide prediction model was developed within a universal kriging framework that incorporated multiple geographic predictors and spatial correlations based on regulatory air quality monitoring data from 2001 to 2019. The regulatory monitoring network, operated by the Korea Environment Corporation, comprised 136 to 422 monitoring sites, depending on the year, with data publicly available on the Air Korea website (https://www.airkorea.or.kr). Geographic predictors represented potential sources of pollution and were derived from over 300 geographical variables related to land use, traffic, population density, and other emission sources, using partial least squares regression [28, 29]. This prediction model yielded moderate to strong predictive performance, as indicated by cross-validation R² values of 0.45 for PM₁₀ and 0.82 for NO₂, values comparable to those reported by previously published nationwide models [30–33]. Given that the pointwise prediction model provides annual average pollutant concentrations at any location in South Korea, we computed air pollution concentrations at the home addresses of all participants for the 1-, 3-, and 6-year period preceding conscription. However, because residential address information was available only at the district level, we first generated annual average pollutant concentrations at all census tract centroids within the study area. We then calculated district-specific, population-weighted averages and assigned these as representative exposure levels to all individuals residing in each district.

For PM_2.5, we applied a ratio-based modeling approach because nationwide regulatory monitoring data have been available only since 2015, in contrast to PM₁₀ and NO₂, which have been monitored since 2001. This model estimates annual average PM_2.5 concentrations for each district by multiplying the ratio of PM_2.5 to PM₁₀ measurements, both of which have been available in Seoul since 2001, by the predicted district-specific PM₁₀ concentrations [27]. The model demonstrated good predictive performance, with an external validation R² of 0.79, and has been widely applied in previous cohort studies on air pollution in South Korea [34–36].

Our primary exposure variable was the district-specific average concentration of air pollutants over the 3 years preceding the physical examination. This time window captures long-term exposure relevant to myopia development while minimizing potential exposure misclassification due to limited information on individual residential history. In South Korea, this 3-year period typically corresponds to the ages of 16 to 18 years, during which most individuals attend high school. Because school district reputation and stability strongly influence university admission prospects, residential mobility during this period is notably low. This socio-educational context supports the assumption of residential stability within districts throughout the exposure period. Estimated pollutant concentrations were analyzed as both continuous and categorical variables (quartiles) to evaluate potential linear and nonlinear dose-response relationships, respectively.

Covariate selection

Potential confounders were selected a priori based on established association with exposure to air pollution and myopia from existing literature and biological plausibility. Individual-level covariates included BMI category and obesity status, which are associated with both residential location (and thus exposure to air pollution) and myopia risk [37]. The year of examination was included to account for temporal trends in air quality and myopia prevalence. Educational attainment was included as a key confounder given its strong association with myopia [37, 38] and its potential correlation with residential location and socioeconomic factors that influence exposure to air pollution. Educational attainment also serves as a proxy for near-work intensity and outdoor time, which are important behavioral determinants of myopia not directly measured in this study.

District-level contextual variables were considered in the secondary analyses to account for neighborhood-level socioeconomic and environmental characteristics that may confound the association between air pollution and myopia. These variables comprised population density (a marker of urbanization), prevalence of atopic dermatitis (a potential shared inflammatory pathway), average household income (socioeconomic status), and the proportion of older residents (demographic composition). However, we recognized that some of these district-level variables might lie in the causal pathway between air pollution and myopia or act as colliders, potentially inducing an overadjustment bias [39]. Therefore, we designated the models that were adjusted only for individual-level covariates in our primary analysis.

Statistical analysis

We explored demographic and district-level characteristics according to the prevalence and severity of myopia and calculated the average concentrations of air pollutants, including PM₁₀, PM₂.₅, and NO₂. We employed multiple logistic regression models to examine the association between long-term exposure to ambient air pollution and the overall prevalence and severity of myopia, including low and high myopia. Odds ratios (ORs) were estimated for each interquartile range (IQR) increase in continuous pollutant concentrations, as well as for each quartile of categorized concentrations, with the first quartile as the reference group.

Health analysis models were developed in three stages: Model 1 was unadjusted. Model 2 was adjusted for individual-level covariates, including BMI category (normal, overweight, obese), year of examination, and educational attainment (high school graduate or less, attending or graduated from a 2- or 3-year college, or attending or graduated from a 4-year university or higher). Model 3 was further adjusted for district-level variables, including population density, prevalence of atopic dermatitis, average household income, and proportion of older adult residents (≥ 65 years) in each residential district. The proportion of older adult residents and population density were obtained from the 2015 Population Census [40, 41]. District-level average monthly household income data (2014–2018) were derived from the National Pension Service eligibility database [42], and the prevalence of atopic dermatitis was obtained from the 2015 Community Health Survey conducted by the Korea Disease Control and Prevention Agency [43]. All district-level variables were categorized into tertiles. Model 2 was used in our primary analysis to minimize the potential overestimation of district-level variables. For sensitivity analysis, we also estimated average exposures during the preceding 1 (PM₁₀, PM_2.5, and NO₂) and 6 years (PM₁₀, NO₂) to evaluate the robustness of the associations. All statistical analyses were conducted using R version 4.4.1 (R Foundation for Statistical Computing).

Results

Overall, 55.3% of participants had myopia, including 17.0% with high myopia. Myopia prevalence varied substantially across demographic, socioeconomic, and regional characteristics (all p < 0.001). A higher prevalence of myopia was observed among residents of Seoul (51.3%), those examined in recent years (53.6% in 2010–2013 vs. 58.0% in 2017–2020), those with higher educational attainment (61.5% in 4-year college graduates vs. 45.9% in those with high school education or below), and those living in districts with a higher population density, greater prevalence of atopic dermatitis, and higher average household income. In contrast, a lower prevalence of myopia was observed among individuals with a higher BMI and those living in areas with larger older populations (Table 1).

Table 1.

Characteristics of the study population and myopia prevalence by subgroups

Variable	Category	Total (N = 167,826)	Myopia
			Any (N = 924,959)	Low (N = 640,655)	High (N = 284,304)
Regions†	Seoul	710,797	415,662 (58.5)	283,138 (39.8)	132,524 (18.6)
	Busan	236,469	129,361 (54.7)	89,556 (37.9)	39,805 (16.8)
	Daegu	184,097	95,671 (52.0)	65,235 (35.4)	30,436 (16.5)
	Incheon	219,164	112,730 (51.4)	79,601 (36.3)	33,129 (15.1)
	Gwangju	117,253	62,717 (53.5)	45,886 (39.1)	16,831 (14.4)
	Daejeon	117,534	62,679 (53.3)	44,538 (37.9)	18,141 (15.4)
	Ulsan	86,512	46,139 (53.3)	32,701 (37.8)	13,438 (15.5)
Period†	2010–2013	672,588	360,467 (53.6)	252,524 (37.5)	107,943 (16.0)
	2014–2016	471,260	258,492 (54.9)	179,853 (38.2)	78,639 (16.7)
	2017–2020	527,978	306,000 (58.0)	208,278 (39.4)	97,722 (18.5)
BMI‡			22.91 ± 4.24	22.87 ± 4.19	23.01 ± 4.34
Obesity†	Normal (< 23)	968,696	538,211 (55.6)	375,548 (38.8)	162,663 (16.8)
	Overweight (23 − 25)	274,871	153,238 (55.7)	106,357 (38.7)	46,881 (17.1)
	Obese (25>=)	428,084	233,510 (54.5)	158,750 (37.1)	74,760 (17.5)
Educational Attainment†	Below High School	358,842	164,874 (45.9)	121,385 (33.8)	43,489 (12.1)
	2-year or 3-year College	381,343	186,599 (48.9)	134,340 (35.2)	52,259 (13.7)
	4-year College or Higher	931,641	573,486 (61.6)	384,930 (41.3)	188,556 (20.2)
Population density†	T1	77,958	39,586 (50.8)	28,340 (36.4)	11,246 (14.4)
	T2	601,675	322,650 (53.6)	227,132 (37.7)	95,518 (15.9)
	T3	992,193	562,723 (56.7)	385,183 (38.8)	177,540 (17.9)
Prevalence of atopic dermatitis†	T1	482,148	259,564 (53.8)	180,048 (37.3)	79,516 (16.5)
	T2	400,783	215,506 (53.8)	150,001 (37.4)	65,505 (16.3)
	T3	788,895	449,889 (57.0)	310,606 (39.4)	139,283 (17.7)
Average monthly income†	T1	616,931	334,890 (54.3)	231,194 (37.5)	103,696 (16.8)
	T2	570,787	310,088 (54.3)	216,142 (37.9)	93,946 (16.5)
	T3	484,108	279,981 (57.8)	193,319 (39.9)	86,662 (17.9)
Proportion of the older population†	T1	464,733	257,775 (55.5)	179,460 (38.6)	78,315 (16.9)
	T2	581,987	326,416 (56.1)	225,694 (38.8)	100,722 (17.3)
	T3	625,106	340,768 (54.5)	235,501 (37.7)	105,267 (16.8)

BMI body mass index, T1 tertile 1 (lowest), T2 tertile 2 (middle), T3 tertile 3 (highest)

† Values are presented as N (%). Percentages in each column represent the myopia prevalence within each characteristic category. Categorical variables are presented as N (%). Between-group differences were tested using the chi-square test (all p < 0.001)

‡ values are presented as mean ± standard deviation; between-group difference was tested by analysis of variance (p < 0.001)

Average concentrations of PM₁₀, NO₂, and PM₂.₅ significantly differed by myopia status and exposure duration (1, 3, and 6 years), with all group differences being statistically significant (p < 0.001). Pollutant levels increased with longer exposure windows, reflecting higher concentrations in earlier years. NO₂ concentrations were consistently higher among individuals with myopia and increased gradually with myopia severity. In contrast, PM₁₀ and PM₂.₅ concentrations tended to decrease with increasing myopia severity relative to the non-myopic group, though these differences were modest (Table 2).

Table 2.

Average concentrations of pollutants according to myopia status

Pollutant	Category	No myopia	Myopia
			All	Low	High
PM10	1 year	46.47 ± 4.17	46.38 ± 4.26	46.44 ± 4.29	46.25 ± 4.20
	3 years	47.77 ± 4.26	47.64 ± 4.29	47.68 ± 4.33	47.53 ± 4.19
	6 years	49.66 ± 4.69	49.42 ± 4.68	49.47 ± 4.71	49.31 ± 4.60
NO2	1 year	25.77 ± 6.24	26.24 ± 6.40	26.19 ± 6.43	26.36 ± 6.32
	3 years	26.48 ± 6.49	26.96 ± 6.64	26.90 ± 6.67	27.08 ± 6.55
	6 years	27.27 ± 6.60	27.74 ± 6.69	27.67 ± 6.72	27.90 ± 6.63
PM2.5	1 year	24.53 ± 1.81	24.44 ± 1.79	24.45 ± 1.78	24.43 ± 1.80
	3 years	24.55 ± 1.16	24.51 ± 1.16	24.51 ± 1.16	24.50 ± 1.15

PM particulate matter, NO₂ nitrogen dioxide

In Model 2, which accounted for individual-level covariates, an IQR increase in PM₁₀ concentrations over the previous 3 years was associated with 4.5% higher odds of overall myopia (OR: 1.045, 95% CI: 1.038–1.051), with stronger associations for low myopia (OR: 1.054, 95% CI: 1.047–1.061) than for high myopia (OR: 1.024, 95% CI: 1.015–1.033). These associations were attenuated in Model 3, which was further adjusted for regional characteristics, and statistical significance was lost for overall myopia. In contrast, NO₂ showed higher and statistically significant differences across all models and myopia types. In Model 2, each IQR increase in NO₂ was associated with higher odds of overall myopia (OR: 1.262, 95% CI: 1.254–1.270), with the strongest effect for high myopia (OR: 1.354, 95% CI: 1.342–1.366). These effects remained significant after adjusting for both individual- and area-level covariates in Model 3 (OR for high myopia: 1.181, 95% CI: 1.162–1.201). PM₂.₅ was inversely associated with all myopia types in both models. In Model 2, the ORs were 0.964 (95% CI: 0.955–0.974) for overall myopia, 0.968 (95% CI: 0.958–0.979) for low myopia, and 0.955 (95% CI: 0.943–0.969) for high myopia; with similar results in Model 3 (Table 3).

Table 3.

Association between 3-year average exposure to PM and NO₂ and the prevalence of myopia: continuous exposure models

Pollutant	Type of Myopia	Model 1	Model 2	Model 3
PM10	Myopia	0.952 (0.948–0.957)	1.045 (1.038–1.051)	1.005 (0.998–1.011)
	Low myopia	0.969 (0.964–0.974)	1.054 (1.047–1.061)	1.021 (1.014–1.029)
	High myopia	0.915 (0.908–0.921)	1.024 (1.015–1.033)	0.967 (0.958–0.976)
NO2	Myopia	1.155 (1.148–1.162)	1.262 (1.254–1.270)	1.146 (1.133–1.159)
	Low myopia	1.136 (1.128–1.143)	1.225 (1.216–1.233)	1.131 (1.117–1.145)
	High myopia	1.201 (1.190–1.211)	1.354 (1.342–1.366)	1.181 (1.162–1.201)
PM2.5	Myopia	0.947 (0.938–0.957)	0.964 (0.955–0.974)	0.948 (0.937–0.959)
	Low myopia	0.954 (0.944–0.964)	0.968 (0.958–0.979)	0.957 (0.945–0.968)
	High myopia	0.934 (0.921–0.946)	0.955 (0.943–0.969)	0.929 (0.915–0.944)

PM particulate matter, NO₂ nitrogen dioxide

* Odds ratios (ORs) and 95% confidence intervals (CIs) estimated per interquartile range (IQR) increase in pollutant concentration

* Model 1: unadjusted

* Model 2: adjusted for individual covariates

* Model 3: adjusted for individual and regional covariates

Using quartile-based exposure variables with Q1 as the reference, PM₁₀ showed weak and positive relationships with overall myopia in Model 2, with slightly elevated ORs in Q2–Q4. These associations diminished in Model 3, and an inverse association was observed for high myopia in Q3 (OR: 0.984; 95% CI: 0.971–0.997) and Q4 (OR: 0.953; 95% CI: 0.938–0.968), consistent with the continuous model. NO₂ demonstrated a consistent dose-response pattern, as in the continuous model. In Model 2, individuals in Q4 had higher odds for overall myopia (OR: 1.356, 95% CI: 1.344–1.369), particularly for high myopia (OR: 1.480, 95% CI: 1.461–1.500), both associations persisted in Model 3 (OR: 1.206, 95% CI: 1.181–1.232). PM₂.₅ was inversely associated with myopia. In Model 2, participants in Q4 had lower odds for overall myopia (OR: 0.918; 95% CI: 0.902–0.935) and high myopia (OR: 0.912; 95% CI: 0.890–0.936), which persisted in Model 3, consistent with the continuous model results (Table 4).

Table 4.

Association between 3-year average exposure to PM and NO₂ and the prevalence of myopia: quartile-based exposure models

Pollutant	Type of Myopia	Exposure Category	Model 1	Model 2	Model 3
PM10	Myopia	Q2	1.002 (0.993–1.010)	1.048 (1.039–1.057)	1.003 (0.995–1.013)
		Q3	0.939 (0.930–0.947)	1.018 (1.009–1.028)	0.968 (0.959–0.977)
		Q4	0.914 (0.906–0.921)	1.049 (1.038–1.060)	0.986 (0.976–0.997)
	Low myopia	Q2	0.996 (0.987–1.006)	1.036 (1.026–1.046)	0.998 (0.989–1.008)
		Q3	0.934 (0.925–0.943)	1.003 (0.993–1.013)	0.961 (0.952–0.971)
		Q4	0.934 (0.925–0.942)	1.054 (1.042–1.066)	1.001 (0.989–1.013)
	High myopia	Q2	1.014 (1.002–1.026)	1.077 (1.064–1.090)	1.015 (1.003–1.028)
		Q3	0.949 (0.938–0.961)	1.055 (1.042–1.068)	0.984 (0.971–0.997)
		Q4	0.869 (0.858–0.880)	1.035 (1.019–1.050)	0.953 (0.938–0.968)
NO2	Myopia	Q2	0.970 (0.962–0.979)	1.032 (1.023–1.041)	0.992 (0.982–1.002)
		Q3	1.029 (1.020–1.038)	1.115 (1.105–1.124)	0.988 (0.976–1.000)
		Q4	1.194 (1.183–1.204)	1.356 (1.344–1.369)	1.144 (1.127–1.160)
	Low myopia	Q2	0.955 (0.946–0.964)	1.007 (0.998–1.017)	0.978 (0.967–0.988)
		Q3	0.987 (0.978–0.997)	1.058 (1.048–1.068)	0.952 (0.939–0.965)
		Q4	1.170 (1.159–1.181)	1.307 (1.294–1.320)	1.118 (1.101–1.136)
	High myopia	Q2	1.008 (0.995–1.020)	1.093 (1.079–1.107)	1.026 (1.011–1.041)
		Q3	1.129 (1.115–1.143)	1.256 (1.240–1.272)	1.076 (1.057–1.096)
		Q4	1.250 (1.235–1.266)	1.480 (1.461–1.500)	1.206 (1.181–1.232)
PM2.5	Myopia	Q2	0.964 (0.946–0.984)	0.984 (0.966–1.002)	0.961 (0.943–0.979)
		Q3	0.949 (0.930–0.968)	0.978 (0.960–0.996)	0.954 (0.936–0.973)
		Q4	0.896 (0.879–0.914)	0.918 (0.902–0.935)	0.908 (0.890–0.926)
	Low myopia	Q2	0.981 (0.962–1.001)	0.981 (0.962–1.001)	0.964 (0.945–0.984)
		Q3	0.978 (0.959–0.998)	0.978 (0.959–0.998)	0.960 (0.941–0.981)
		Q4	0.921 (0.903–0.939)	0.921 (0.903–0.939)	0.912 (0.893–0.932)
	High myopia	Q2	0.964 (0.940–0.988)	0.990 (0.966–1.015)	0.955 (0.931–0.979)
		Q3	0.930 (0.907–0.954)	0.976 (0.951–1.001)	0.942 (0.918–0.967)
		Q4	0.876 (0.854–0.898)	0.912 (0.890–0.936)	0.898 (0.874–0.923)

PM particulate matter, NO₂ nitrogen dioxide

* Odds ratios (ORs) and 95% confidence intervals (CIs) comparing higher quartiles (Q2–Q4) with the lowest quartile (Q1, reference group) of pollutant exposure

* Model 1: unadjusted

* Model 2: adjusted for individual covariates

* Model 3: adjusted for individual and regional covariates

Sensitivity analyses using alternative exposure windows demonstrated the robustness of these associations. When exposure was defined based on a 1-year average prior to the examination, the results for NO₂ were consistent with those observed using the 3-year exposure window, showing persistent positive associations across all myopia categories. Conversely, the statistically significant association observed between PM₁₀ and high myopia in the 3-year model was no longer evident in the 1-year model (Supplementary Tables S1 and S2). When exposure was extended to a 6-year average, both PM₁₀ and NO₂ showed results that were largely consistent with those of the 3-year exposure model, particularly regarding the direction and significance of associations (Supplementary Tables S3 and S4). In temporal sensitivity analysis restricting PM₁₀ and NO₂ to the 2018–2020 period (the only years with PM_2.5 data available), PM₁₀ demonstrated an inverse association with myopia (OR, 0.957; 95% CI, 0.940–0.974), similar to the pattern observed for PM_2.5. In contrast, NO₂ maintained positive associations (OR: 1.239; 95% CI: 1.219–1.259) during this period (Supplementary Tables S5 and S6).

Discussion

In this nationwide study of South Korean military conscription participants, we observed a significant association between ambient air pollution and the prevalence of myopia. NO₂ consistently demonstrated the strongest and most robust associations with myopia, particularly high myopia, regardless of whether exposure was treated as a continuous variable or categorized into quartiles. PM₁₀ was also positively associated with myopia, although the effects were weaker and less consistent, showing slightly stronger associations with low myopia. In contrast, PM_2.5 showed a consistent inverse association across all myopia categories. These findings suggest that among the pollutants examined, NO₂ exposure may play a more prominent role in the development or progression of myopia, particularly high myopia.

The observed association with NO₂ aligns with findings from previous studies. For instance, a retrospective cohort study in Taiwan involving approximately 97,000 children reported that long-term exposure to PM₂.₅ and NO₂ was associated with an increased risk of myopia [18]. Our results further suggest that air pollution exposure may contribute not only to the onset of myopia but also to its progression to high myopia. Supporting this biological plausibility, Lasagni et al. [44] demonstrated that chronic exposure to urban air pollution induces oxidative stress, inflammation, and epithelial hyperplasia in mouse corneas, which could potentially affect ocular development and refractive outcomes. Although not focused specifically on myopia, a meta-analysis examining the relationship between air pollution and visual impairment found that higher exposure to NO₂ was associated with an increased risk of visual dysfunction, with children and adolescents appearing particularly vulnerable [45]. In our study, participants were 19-year-old males who had recently graduated from high school, suggesting that their exposure to ambient air pollution during adolescence—a critical developmental period—may be particularly relevant to the observed associations.

We found that NO₂ exhibited a stronger association with myopia than PM₁₀, with particularly pronounced effects for high myopia. NO₂ is a marker of traffic-related pollution (vehicle emissions) and is highly correlated with dense urban environments [46]. Several mechanisms may explain this association. First, children living in high-NO₂ urban centers may spend less time outdoors, either because of concerns about air quality or lack of accessible green space, thereby forfeiting the protective effects of outdoor light exposure [11, 47]. NO₂ exposure may therefore directly affect ocular development through oxidative stress and inflammation. Documented NO₂-associated ocular surface changes include inflammation, tear film instability, and increased dry eye syndrome risk [48, 49]. Novaes et al. [50] reported a dose-dependent increase in the conjunctival goblet cell density following NO₂ exposure, indicating a chronic inflammatory response. Chronic ocular surface inflammation may affect axial elongation and refractive development during critical growth periods. Second, behavioral pathways may mediate the effects; children in high-pollution areas may spend less time outdoors owing to air quality concerns or reduced access to green spaces, thereby limiting protective outdoor light exposure [11, 47]. Third, NO₂ may be a marker for broader urbanization-related risk factors, although our findings remained significant after adjusting for population density and socioeconomic indicators, suggesting effects beyond merely serving as an urbanization proxy.

Educational attainment was strongly associated with myopia in our cohort, which is consistent with the existing literature demonstrating that educational level and reduced outdoor time are major contributors to myopia development [37, 38, 51]. Importantly, air pollution associations persisted even after adjusting for educational attainment and other sociodemographic factors, suggesting independent contributions. However, residual confounding by unmeasured individual-level behaviors, particularly variations in outdoor time and near-work intensity that could not be fully captured by educational attainment, cannot be ruled out. Future studies with detailed individual-level behavioral data would help further separate air pollution effects from lifestyle factors. After adjusting for regional-level covariates, the effect of NO₂ was considerably attenuated, and the association with PM₁₀ was no longer statistically significant. One plausible explanation is overadjustment bias, which can occur when controlling for variables on the causal pathway or acting as colliders [39]. Specifically, some district-level variables, such as population density and average income, may serve as intermediaries or colliders in the causal structure. These variables are influenced by urbanization, which is often associated with higher levels of air pollution, and are also independently related to myopia risk. Moreover, in our study, both exposure and regional-level covariates were measured at the same administrative level (city or county), increasing the likelihood that these variables reflected shared spatial structures. Certain socioenvironmental features captured in district-level statistics may be influenced by the long-term presence of air pollution in an area. Thus, by conditioning on such variables, we may have inadvertently blocked part of the causal effect or introduced spurious associations, potentially biasing the estimates toward the null [39, 52]. Although regional-level adjustments can mitigate contextual confounding factors, they may also obscure the total effect of exposure in a semi-ecological study setting, particularly when exposure and covariates share spatial resolution and structure. For this reason, we treated Model 2, which was adjusted only for individual-level covariates, as our primary model to better reflect the total association between air pollution and myopia without introducing potential bias from conditioning on regional-level variables.

Another important consideration is exposure assessment and the spatial resolution of the data. We assigned pollution exposure at the district level (averaged over a relatively large area), which undoubtedly introduced exposure misclassification, as individuals within the same district may have experienced substantially different pollution levels. For example, someone residing near a busy roadway may have higher NO₂ exposure than someone in a quieter residential area of the same district, yet both would be assigned the same district-level average. To minimize potential bias, we excluded non-metropolitan regions from our analysis. In 2020, the median district area in non-metropolitan regions was 519 km² compared to 36 km² in metropolitan areas [53]. The larger spatial units in non-metropolitan areas make it more difficult to assign representative exposure values, which could further amplify exposure misclassification and undermine the validity of our estimates. As a further step toward reducing exposure misclassification, our simulation study demonstrated that population-representative average concentrations estimated using the universal Kriging model, applied in the present study, yield less biased health effect estimates than simple area averages or alternative model-based estimates when address data are incomplete [54].

Regarding outcome measurements, our study employed non-cycloplegic autorefraction, which may have overestimated the prevalence of myopia compared with cycloplegic methods [25]. However, this measurement approach does not undermine our findings regarding the harmful effects of NO₂ and PM₁₀ on myopia risk. Because accommodation-related measurement errors occur independently of air pollution exposure levels, any misclassification of myopia status would be non-differential with respect to NO₂ and PM₁₀ exposure. Non-differential misclassification typically attenuates associations toward the null, rather than creating spurious relationships [52]. Therefore, our observed associations between air pollutants and increased myopia prevalence likely represent conservative estimates, and the actual harmful effects of NO₂ and PM₁₀ on ocular health may be more potent than reported here.

The inverse association observed for PM₂.₅ warrants careful interpretation. Our temporal sensitivity analysis revealed that PM₁₀ also demonstrated an inverse association when restricted to the 2018–2020 period, whereas NO₂ maintained a positive association (Supplementary Tables S5 and S6). This pattern suggests that the inverse PM₂.₅ association reflects temporal confounding or cohort effects rather than a true protective effect.

This study has some limitations. First, its cross-sectional design precludes establishment of temporal and causal inferences. Although we estimated long-term exposure by averaging pollutant concentrations over several years, we did not follow individuals from childhood to directly observe the incidence or progression of myopia. Because our dataset lacked information on age of myopia onset, the observed associations may reflect effects on myopia progression during late adolescence rather than initial development during early childhood. Nonetheless, given that myopia typically develops during the school years, our exposure window during late adolescence likely captured the relevant chronic exposures. Second, we lacked residential history data, which may have resulted in exposure misclassification. District-level exposure assignments may introduce differential misclassification, with potentially larger measurement errors in heterogeneous high-pollution urban districts than in more homogeneous low-pollution areas, potentially biasing associations in either direction. Third, individual-level data on important confounders, including time spent outdoors, near-work behaviors, and parental myopia, which strongly affect myopia development, were unavailable. While educational attainment has been used as a proxy indicator for near-work intensity and outdoor activity, this approach does not fully capture individual behavioral variability or genetic predisposition. Accordingly, residual confounding owing to unmeasured variables cannot be ruled out. Fourth, our male-only sample limits generalizability to females, although no clear biological basis exists to predict sex-specific differences in the effects of air pollution on myopia. Finally, PM₂.₅ analysis was restricted to 2018–2020 examinations owing to monitoring availability since 2015, limiting comparability with PM₁₀ and NO₂ analyses and potentially explaining the inverse PM₂.₅ association observed.

This study contributes to the growing body of evidence linking ambient air pollution to myopia, with novel emphasis on high myopia as a pollution-sensitive outcome. Contrary to previous assumptions that high myopia is primarily genetically determined, our findings demonstrated strong associations between air pollution, particularly NO₂, and prevalence of high myopia. This suggests that air pollution may contribute not only to the onset of myopia but also to its progression to more severe forms. The stronger associations observed in high myopia may reflect the cumulative impact of chronic air pollution exposure on ocular development through oxidative stress and inflammatory pathways that could accelerate axial elongation. These findings broaden the health impact perspective on air pollution beyond the well-established respiratory and cardiovascular effects to include visual health, which is an emerging public health concern, particularly in rapidly urbanizing East Asian regions that experience both high air pollution and increased myopia prevalence. The ubiquity of air pollution exposure, combined with evidence that poor air quality may discourage protective outdoor activities in urban areas, underscores the importance of addressing environmental factors in comprehensive myopia prevention strategies.

Conclusions

This extensive, nationally representative study of young Korean men found that long-term exposure to ambient PM₁₀ and NO₂ was associated with higher myopia prevalence, with associations attenuated after adjustment for regional contextual factors. The association with NO₂, a marker of traffic-related air pollution, was particularly strong and more pronounced for high myopia. These findings highlight the potential relevance of ambient air quality to visual health, which is an emerging public health issue in East Asia. Although causality could not be inferred from the cross-sectional design and district-level exposure assignments, the consistency and strength of the observed associations warrant further investigation. Longitudinal studies with individual-level exposure assessments and repeated refractive error measurements are needed to clarify temporal relationships and identify critical windows of susceptibility. Policies aimed at improving air quality may yield additional benefits for ocular health, reinforcing the broader need for environmental action to promote population well-being.

Abbreviations

PM

Particulate matter

NO₂

Nitrogen dioxide

OR

Odds ratio

IQR

Interquartile range

CI

Confidence interval

BMI

Body mass index

SE

Spherical equivalent

D

Diopters

Authors’ contributions

K.K: Writing—original draft, methodology, investigation, conceptualization, data curation, formal analysis, funding acquisition, and visualization. S.Y.K: Writing—review & editing, methodology, investigation, validation, and formal analysis. S.Y: Writing—review & editing, conceptualization, project administration, resources, and supervision. All authors read and approved the final manuscript.

Funding

This study was supported by the Future Medicine Pioneer Research Grant (K2211901) from Korea University, Seoul, Republic of Korea. The funder had no role in the study design; collection, analysis, or interpretation of data; writing of the report; or the decision to submit the article for publication.

Data availability

The data that support the findings of this study are derived from the South Korean military conscription examination database. Access to these data is restricted and requires formal approval through institutional and governmental review processes. Therefore, the data are not publicly available but can be accessed by authorized researchers upon reasonable request and approval from the relevant authorities.

Declarations

Ethics approval and consent to participate

This study adhered to the principles outlined in the Declaration of Helsinki and was approved by the Institutional Review Board of the Armed Forces Medical Command of the Republic of Korea (IRB no: AFMC-19081-IRB-19-057), which waived the need for written informed consent given that only de-identified data collected by a government agency were analyzed and no identifiable information was accessed.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.