Mortality burden attributable to long-term exposure to fine particulate matter among older adults in Korea

Study population

We targeted 5,562,993 adults aged 65 years and older in 2010. To construct a retrospective cohort of senior adults, we utilized demographic variables, health insurance claim data, and vital status from the National Health Information Database (NHID; NHIS-2021-1-694). Because health insurance coverage is mandatory for individuals legally residing in Korea, NHID provides comprehensive data for almost the entire population [16]. NHID data span national, provincial, and municipal levels, effectively covering nearly all residents in Korea [17]. NHID records, linked to death registry data provided by Statistics Korea, were used to identify causes of death among older adults. Among older adults in 2010, we excluded subjects with missing demographic information, such as gender, age, income level, and area of residence (n=201,804, 3.63%), as well as subjects with death records prior to 2010 (n=1,157, 0.02%). To avoid bias introduced by increased mortality during the coronavirus disease 2019 pandemic, the study period was restricted to January 2010 through December 2019. During the follow-up period, deaths from causes not of interest in our analysis were censored.

Outcome definition

We defined the major causes of death related to long-term exposure to PM_2.5 as primary health outcomes. Causes of death were selected based on previous studies demonstrating associations with PM_2.5 exposure [3,4,6,7,11,12,18-21] (Supplementary Material 1). Data on the selected causes of death and its International Classification of Diseases 10th revision (ICD-10) codes are as follows: IHD (I20-I25), stroke (I60-I69), ALRI (J12-J18, J20-J22), COPD (J40-J44), LC (C33-C34), and T2DM (E11).

Particulate matter with an aerodynamic diameter <2.5 μm data

The PM_2.5 exposure data for administrative districts were predicted using the Community Multiscale Air Quality model (CMAQ, version 4.7.1), with a spatial resolution of 9 km×9 km. Since nationwide PM_2.5 monitoring in Korea was not implemented before 2015, we relied on modeled PM_2.5 data. Modeled PM_2.5 concentrations employed in this study have been widely used in environmental epidemiological studies conducted in Korea [3,22-24]. The CMAQ model demonstrated high correlation with monitored data from 2010 to 2019 (Supplementary Material 2). Monthly correlations between monitored and modeled PM_2.5 concentrations across cities and provinces ranged from 0.878 to 0.999 during 2015-2019. A detailed description of the CMAQ model is provided in Supplementary Material 3 [3,22-24]. We linked 12-month moving average PM_2.5 exposure values to individuals based on their residential district (250 administrative districts). Because residential information was updated annually, exposure assignments were revised each calendar year to account for residential relocation.

Covariates

Individual-level covariates included gender, income level at baseline, insurance enrollment type (self-employed or employee subscriber), area of residence (7 metropolitan cities and 9 provinces), and underlying diseases at baseline (2010). Income levels, segmented into 20 quantiles, were categorized into 6 groups: Medical Aid recipients (category 0), very low (1-4), low (5-8), moderate (9-12), high (13-16), and very high (17-20). The Charlson comorbidity index (CCI) was calculated using health insurance claims data from NHID for the year preceding the baseline (January-December 2009). Underlying diseases were determined based on the CCI, with a CCI score of 1 or higher classified as “yes,” and a score of 0 classified as “no.”

In this study, individual-level potential confounders such as smoking, alcohol consumption, and physical activity could not be directly adjusted due to their absence in claims data. Therefore, we used district-level confounders, as suggested in previous studies. We obtained the population, percentage of the population aged 65 years and older (%), education level (i.e., the proportion of individuals with high school diplomas [%]), and smoking rate (%) from the Korean Statistical Information Service. The annual mean temperature (°C) and rainfall (mm) were obtained from the Korea Meteorological Administration. All district level variables were adjusted as time-varying covariates to reflect temporal changes (i.e., yearly). For the proportion of older adults with high school diplomas, data were collected at 5-year intervals in 2010 and 2015. Therefore, 2010 data were assigned for the period 2010-2014, and 2015 data were assigned for 2015-2019.

Associations between long-term exposure to particulate matter with an aerodynamic diameter <2.5 μm and cause-specific mortality

To evaluate the association between long-term exposure to PM_2.5 and cause-specific mortality, we employed a time-varying Cox proportional hazard model. The time variable was defined in monthly intervals (from 1 to 120 months). Biological age was used as the time scale to examine age-dependent associations between long-term PM_2.5 exposure and cause-specific mortality [25]. We constructed 4 models: Model 1: unadjusted; Model 2: adjusted for gender, type of insurance enrollment, income level, calendar year, and residence area (using strata terms); Model 3: adjusted for all covariates from model 2 and underlying diseases; Model 4: adjusted for all covariates from model 3 and district-level confounders. The model with the lowest Akaike information criterion (model 4) was selected as the main model (Supplementary Material 4).

To investigate the exposure-response relationship between the 12-month moving average PM_2.5 and cause-specific mortality risk, we tested for non-linearity using restricted cubic spline models with varying knots (3, 4, and 5). Among these models, the spline with 3 knots had the lowest Akaike information criterion and was thus selected.

We performed subgroup analyses by gender, age group (young-old [65-74 years] vs. old-old [≥75 years]) [26], and Medical Aid status (Medical Aid beneficiaries vs. non-Medical Aid beneficiaries). A t-test was conducted to assess differences in effect size (log hazard ratio [HR]) between the 2 groups in each subgroup analysis, calculated as follows:

where β indicates the effect size, and SE indicates the standard error for each effect size.

Four sensitivity analyses were conducted. First, we analyzed the association between 24-month, 36-month, and 48-month moving average PM_2.5 exposure and cause-specific mortality. Second, we performed the same analysis after excluding individuals who died within 1 month, 2 months, and 3 months from the baseline. Third, we performed the analysis on individuals who did not move between the 16 regions during the study period. Fourth, we constructed a two-pollutant model by adjusting for sulfur dioxide (SO₂), nitrogen dioxide (NO₂), and ozone (O₃) to estimate the effect size of PM_2.5. All estimated HRs with 95% confidence intervals (CIs) were presented per 10 μg/m³ increase in PM_2.5 concentration.

Mortality burden attributable to particulate matter with an aerodynamic diameter <2.5 μm

We estimated the mortality burden attributable to long-term exposure to PM_2.5 for 5 specific causes (IHD, stroke, ALRI, COPD, and T2DM) from 2010 to 2019. Because our study did not identify a positive association for LC, we did not estimate excess deaths for LC. Excess mortality rates per 100,000 older adults were also calculated to facilitate comparisons across specific causes. Excess deaths were estimated using the equation:

where “Deaths” represents the number of deaths due to a specific cause among individuals aged 65 years or older, β indicates the log HR from the main model (model 4) in this study. ΔPM_2.5 indicates the difference in the concentration of PM_2.5 between the current level and a reference level, which is assumed to be the baseline concentration for comparison. In other words, if the current concentration is higher than the reference level, the difference represents the health impact that can be attributed to that level of pollution. We set the reference concentration to 5 μg/m³ according to the 2021 WHO air quality guideline [27]. Excess deaths were computed using annual mortality and PM_2.5 concentration data. The 95% CIs for excess deaths were based on the 95% lower and upper bounds of the HR associated with changes in PM_2.5 concentrations. Excess mortality rates were standardized per 100,000 population. Our estimates were also compared to mortality burdens reported in recent systematic reviews [4]. Since no significant associations were observed for LC and T2DM, excess mortality was estimated only for IHD, stroke, ALRI, and COPD.

Data preprocessing, statistical analyses, and visualization were conducted using SAS version 9.4 (SAS Institute Inc., Cary, NC, USA) and R version 4.2.1 (R Foundation for Statistical Computing, Vienna, Austria). Statistical significance was set at a p-value <0.05.

Ethics statement

This study was exempt from review by the Institutional Review Board of Seoul National University Hospital, Korea (IRB No. E-2105-043-1218). Since personal information managed by NHIS was not accessed by the authors, ethical approval was not required.

Long-term exposure to particulate matter with an aerodynamic diameter <2.5 μm and cause-specific mortalities

This study used a health-insurance-based retrospective cohort to follow the entire Korean population of older adults from 2010, evaluating associations between long-term PM_2.5 exposure and 6 cause-specific mortalities. Our findings indicate that long-term PM_2.5 exposure was associated with increased mortality from IHD, stroke, ALRI, COPD, and T2DM. The association between PM_2.5 and IHD mortality was more pronounced in women than in men. Furthermore, associations of PM_2.5 with COPD and T2DM were more clearly observed among the old-old population (≥75 years) compared to the young-old population (65-74 years).

We estimated the mortality burden attributable to long-term PM_2.5 exposure. Due to population aging, the number of deaths attributable to long-term PM_2.5 exposure increased from 4,888 in 2010 to 5,179 in 2019.

Comparison with previous studies

This study estimated effect sizes between long-term PM_2.5 exposure and cause-specific mortality, assuming linearity. We identified associations between long-term PM_2.5 exposure and cause-specific mortality (IHD, stroke, ALRI, COPD, and T2DM), but not with LC. These findings are consistent with previous studies [1,3,4,11,12,15,28] (Supplementary Material 16); however, our estimated effect sizes were relatively lower compared to those in prior studies. We speculate this may be due to differences in regional characteristics, the study period, improvements in health conditions, and the implementation of PM_2.5 reduction measures. Specifically, studies vary by countries or regions, population demographics (older adults vs. general population), and spatial scales (national, multinational, or specific metropolitan areas), influencing ambient PM_2.5 concentration levels and distributions. For instance, Orellano et al. [15], adopted a global approach, encompassing various countries with diverse air pollution levels, whereas our study focused specifically on older adults across Korea. Differences in PM_2.5 exposure levels, healthcare systems, policies, and underlying health conditions could therefore explain variations in effect sizes. Moreover, a previous study reported notable mortality increases associated with air pollution in 7 major Korean cities [3]. However, that study was restricted to urban areas, limiting its generalizability due to variations in PM_2.5 levels, composition, and socioeconomic factors across regions.

Gender differences

We found older women to be more susceptible to the adverse effects of long-term PM_2.5 exposure, especially mortality from IHD, COPD, and T2DM. These results align with previous research [11,12,29]. Hormonal changes, such as the decline in estrogen levels after menopause, diminish estrogen’s cardiovascular protective effects, leading to impaired endothelial function, increased systemic inflammation, and heightened atherosclerosis risk [30]. Additionally, older women might have greater metabolic sensitivity, including increased oxidative stress and susceptibility to metabolic syndrome [31,32]. Differences in vascular structure, particularly the greater susceptibility of women to microvascular dysfunction, may further amplify the cardiovascular effects of PM_2.5 exposure [33]. Exposure to air pollution promotes oxidative stress by increasing reactive oxygen species, which might lead to faster depletion of antioxidant defenses in women, exacerbating vascular and metabolic dysfunction [33].

Regarding COPD, non-infectious factors such as air pollution, smoking, and environmental exposures are primary triggers for disease onset [34]. Women typically have smaller lung capacities and narrower airways, resulting in higher particulate deposition relative to lung tissue volume, potentially intensifying the respiratory impact of air pollution [35]. Furthermore, lung function declines may progress more rapidly in women, increasing their susceptibility to COPD [36]. Post-menopausal hormonal changes might also influence systemic inflammation and respiratory health, further heightening COPD risk [37]. Psychosocial stress may also contribute, as older women generally report higher mental distress than older men [38], and stress is recognized as a potential risk factor for COPD [39].

Age differences

Previous studies from the United States and Hong Kong reported stronger associations between long-term PM_2.5 exposure and mortality from IHD and COPD in young-old adults [11,12]. In contrast, we found that older groups (≥75 years) were more susceptible to adverse PM_2.5 effects, particularly for IHD, COPD, and T2DM mortality, compared to young-old adults (65-74 years). This discrepancy might be explained by several factors. Aging correlates with declines in lung and cardiovascular functions and reduced ability to cope with oxidative stress and inflammation, potentially increasing vulnerability to PM_2.5’s harmful effects [40]. Variations in PM_2.5 concentrations, lifestyles, regional differences, and ethnicity may also explain these discrepancies. For example, compared with the United States and Hong Kong, Korea experiences higher PM_2.5 levels, potentially contributing to observed differences in susceptibility and health outcomes.

Mortality burden due to long-term exposure to particulate matter with an aerodynamic diameter <2.5 μm

The estimated number of excess deaths and mortality rate from ALRI attributable to PM_2.5 steadily increased from 2010 to 2019, with 2019 exhibiting the highest mortality burden among evaluated diseases (Figure 2). This trend likely reflects the rising proportion of older adults, given that respiratory diseases represent a leading cause of death in this demographic [41,42].

Between 2010 and 2015, the mortality rate from excess deaths due to PM_2.5 exposure increased but declined from 2015 onward. The observed reduction from 2015 to 2019 is attributable to decreased mortality rates from IHD, stroke, COPD, and T2DM among individuals aged 65 and older. Advances in healthcare have improved disease management and reduced mortality, even in patients with these chronic conditions, thereby increasing life expectancy and decreasing death rates. However, ALRI-related mortality rates have increased, likely due to complications such as pneumonia in bedridden patients in long-term care facilities, where ALRI is frequently listed as the primary cause of death in Korea. Given the continuing population aging, the mortality burden from these 5 major diseases (IHD, stroke, ALRI, COPD, and T2DM) is unlikely to decrease substantially in the future. Consequently, efforts should focus on mitigating air pollution exposure. Preventive measures, including source control of particulate matter pollution and individual-level interventions (e.g., masks or air purifiers), could be critical for mitigating the long-term health impacts associated with PM_2.5 exposure.

Lung cancer mortality

Previous studies have reported significant associations between long-term PM_2.5 exposure and LC mortality [3,4,15,28]. However, our study found a statistically significant but marginal negative association (HR, 0.972; 95% CI, 0.948 to 0.996), which could be attributed to several factors. First, detecting significant associations between long-term PM_2.5 exposure and mortality may require extended observation periods. Since our study focused on associations using a 12-month moving average PM_2.5 exposure, cumulative exposure might have been insufficiently captured. Additionally, our older study population may have included many advanced-stage LC cases, making it difficult to detect short-term effects of recent PM_2.5 exposure.

Sensitivity analyses

The results of our sensitivity analyses were generally consistent with those of the main analysis. When individuals who died within 1 month to 3 months of baseline were excluded, overall HRs decreased slightly, but statistically significant associations persisted for IHD, ALRI, and COPD (Supplementary Material 12). However, associations for stroke and T2DM lost statistical significance. This suggests the observed effects of long-term PM_2.5 exposure remain robust even after excluding vulnerable populations. Conversely, failing to account for residential mobility slightly exaggerated effect sizes for all causes of death (Supplementary Material 13). This might be due to overestimating exposure levels when residential changes are ignored. Therefore, incorporating residential mobility is important for enhancing the accuracy of exposure assessments. Additionally, among older adults, shorter exposure windows (e.g., 1-year moving averages) were associated with higher HRs for IHD, stroke, ALRI, and COPD, but not LC or T2DM mortality (Supplementary Material 11). The reasons underlying this intriguing finding remain unclear but may involve factors such as underlying comorbidities or lifestyle changes. Future studies should explore these potential explanations more thoroughly. These results highlight the importance of considering residential mobility and additional demographic and environmental factors in future studies to comprehensively assess the health impacts of PM_2.5 exposure.

Strengths and limitations

To the best of our knowledge, this is the first study to evaluate associations between long-term PM_2.5 exposure and cause-specific mortality in nearly the entire population of older adults in Korea. Comprehensive subgroup and sensitivity analyses enhance the robustness of our findings. Our results can be generalized to other countries with similar PM_2.5 exposure levels and demographic characteristics. These findings provide valuable insights into the mortality burden among older adults attributable to long-term PM_2.5 exposure, supporting policy formulation aimed at reducing PM_2.5 levels and implementing appropriate interventions (e.g., wearing masks). Additionally, this study not only estimates HRs for cause-specific mortality but also quantifies excess deaths due to long-term PM_2.5 exposure among older Koreans. Our estimates of excess mortality are substantially lower than global estimates provided by the Institute for Health Metrics and Evaluation using the Global Burden of Disease 2021 data. Specifically, excess death estimates from global studies are approximately 1.3 times to over 9.3 times higher than those reported in our study (Supplementary Material 17). This discrepancy suggests that global assessments, while useful for illustrating general disease burdens from air pollution, likely overestimate mortality due to inherent uncertainties. Therefore, our results offer a more precise estimation of the PM_2.5-related mortality burden, informing targeted public health interventions and policies.

Nonetheless, this study has several limitations. First, we did not consider individual-level potential confounders, such as smoking, drinking, and physical activity status [43-45]. These factors could be potential confounders in the relationship between long-term exposure to PM_2.5 and cause-specific mortality [46]. Additionally, residual confounding from unmeasured factors cannot be completely excluded. However, we minimized these issues by adjusting for district-level indicators as recommended by previous research [47]. Furthermore, the significantly lower participation rate of older Koreans in NHIS health screening compared to the general adult population limited our use of health screening data. We deliberately avoided incorporating such data to prevent selection bias, given potential differences between older participants and non-participants. Second, exposure measurement bias and misclassification may have occurred [48]. PM_2.5 exposures in this study were assigned using the CMAQ model at a 9 km×9 km grid resolution, potentially failing to capture finer-scale individual exposure variations. This spatial limitation might introduce measurement error, particularly in environmentally heterogeneous areas. Future research should consider employing higher-resolution exposure data (e.g., 1×1 km grids) and integrating individual-level exposure metrics (e.g., time-varying patterns and agent-based modeling) to improve exposure assessment precision and better clarify associations between long-term air pollution exposure and health outcomes [49].

These measurement errors could attenuate observed associations, biasing results toward null findings and obscuring the true relationship between exposure and mortality outcomes.

Our findings elucidate the associations between long-term PM_2.5 exposure and cause-specific mortality in Korea’s population of older adults from 2010 to 2019. We quantified the mortality burden attributable to long-term PM_2.5 exposure among older Korean adults, providing crucial insights for targeted public health interventions.