Projecting Lifetime Health Outcomes and Costs Associated with the Ambient Fine Particulate Matter Exposure among Adult Women in Korea

We sought to estimate the lifetime healthcare costs and outcomes associated with the exposure to the escalated concentration of fine particulate matter (particle size < 2.5 μm, PM2.5) among adult Korean women. We adapted a previously developed Markov model, and a hypothetical cohort composed of Korean women was exposed to either a standard (15 μg/m3) or increased (25 μg/m3) concentration of PM2.5. The time horizon of the analysis was 60 years, and the cycle length was 1 year. The outcomes were presented as direct healthcare costs and quality-adjusted life years (QALYs), and costs were discounted annually at 5%. Deterministic and probabilistic sensitivity analyses were performed. The model estimated that when the exposure concentration was increased by 10 μg/m3, the lifetime healthcare cost increased by USD 9309, which is an 11.3% increase compared to the standard concentration group. Women exposed to a higher concentration of PM2.5 were predicted to live 30.64 QALYs, compared to 32.08 QALYs for women who were exposed to the standard concentration of PM2.5. The tendency of a higher cost and shorter QALYs at increased exposure was consistent across a broad range of sensitivity analyses. The negative impact of PM2.5 was higher on cost than on QALYs and accelerated as the exposure time increased, emphasizing the importance of early intervention.


Introduction
Air pollution represents one of the biggest environmental risks to health [1]. In 2012, more than three million deaths were attributable to ambient air pollution [2]. Among the pollutants, particulate matter with the diameter less than 2.5 µm (PM 2.5 ) is known to be associated with the increased morbidity and mortality of various diseases [3]. PM 2.5 penetrates within the respiratory tract and circulates in the blood stream due to its small size. As a result, PM 2.5 affects not only the respiratory system but also the cardiovascular system and can cause various health problems. In fact, PM 2.5 was the fifth leading cause of death worldwide following high blood pressure and smoking [4].
The World Health Organization (WHO) Air Quality Guidelines (AQGs) has recommended an annual average PM 2.5 concentration of 10 µg/m 3 as the target value and three interim targets (IT; IT-1 35 µg/m 3 , IT-2 25 µg/m 3 , IT-3 15 µg/m 3 ), which have been shown to be achievable with successive and sustained abatement measures [5]. These concentrations were chosen based on the significance of their effect on survival, where the AQG target value was the lowest level at which the total, cardiopulmonary, and lung cancer mortality were shown to increase with more than 95% confidence in response to PM 2.5 in the American Cancer Society study. In South Korea, the PM 2.5 level has been continually maintained over the guideline value since 2015 when the South Korean government started the official observation of PM 2.5 [6]. However, due to geographical and seasonal reasons, the airborne fine particulate matters from foreign high-emission areas add to the burden of domestic pollution [7]. Therefore, despite the efforts of the government to manage the annual concentration, the potential negative health effects of PM 2.5 have been of great concern among the Korean people.
Though there have been many studies reporting an increased disease risk associated with the elevated PM 2.5 [8][9][10][11], studies projecting the lifetime economic effect of diseases due to PM 2.5 are limited. In an economic burden of disease study performed in 111 cities, the total economic cost caused by particulate matter pollution in 2004 was estimated to be approximately USD 29,178.7 million [12]. If the concentration of PM 2.5 decreases by 10 µg/m 3 , more than USD 22 million of economic benefit will occur annually in Seoul, South Korea [13]. Moreover, life expectancy will be lengthened by 0.35 years [14]. However, there have been no studies estimating the lifetime cost and a much lower quality of life, which is reported to be significantly affected by PM 2.5 exposure [15,16]. The purpose of this study was to estimate the economic and health outcomes of ambient PM 2.5 exposure for a lifetime among Korean women.

Markov Model
We have adopted a previously developed Markov state-transition model to simulate the natural history of PM 2.5 exposure ( Figure 1) [17]. At the start of the model, the cohort was exposed to either standard or increased concentration of PM 2.5 . Subsequent Markov model pathways were same with four diseases and their corresponding health states. The whole cohort started from event-free health state and transferred to each health state according to the probabilities derived from reference data. All health states were mutually exclusive and collectively exhaustive so the patients could be assigned to only one health state at any given time [18,19].
started the official observation of PM2.5 [6]. However, due to geographical and seasonal reasons, the airborne fine particulate matters from foreign high-emission areas add to the burden of domestic pollution [7]. Therefore, despite the efforts of the government to manage the annual concentration, the potential negative health effects of PM2.5 have been of great concern among the Korean people.
Though there have been many studies reporting an increased disease risk associated with the elevated PM2.5 [8][9][10][11], studies projecting the lifetime economic effect of diseases due to PM2.5 are limited. In an economic burden of disease study performed in 111 cities, the total economic cost caused by particulate matter pollution in 2004 was estimated to be approximately USD 29,178.7 million [12]. If the concentration of PM2.5 decreases by 10 μg/m 3 , more than USD 22 million of economic benefit will occur annually in Seoul, South Korea [13]. Moreover, life expectancy will be lengthened by 0.35 years [14]. However, there have been no studies estimating the lifetime cost and a much lower quality of life, which is reported to be significantly affected by PM2.5 exposure [15,16]. The purpose of this study was to estimate the economic and health outcomes of ambient PM2.5 exposure for a lifetime among Korean women.

Markov Model
We have adopted a previously developed Markov state-transition model to simulate the natural history of PM2.5 exposure (Figure 1) [17]. At the start of the model, the cohort was exposed to either standard or increased concentration of PM2.5. Subsequent Markov model pathways were same with four diseases and their corresponding health states. The whole cohort started from event-free health state and transferred to each health state according to the probabilities derived from reference data. All health states were mutually exclusive and collectively exhaustive so the patients could be assigned to only one health state at any given time [18,19]. The four diseases, lung cancer, myocardial infarction, stroke, and COPD, were selected based on previous systematic reviews and official reports, in which those diseases were named as some of the most affected diseases due to PM2.5 exposure [1,[20][21][22][23]. Of The four diseases, lung cancer, myocardial infarction, stroke, and COPD, were selected based on previous systematic reviews and official reports, in which those diseases were named as some of the most affected diseases due to PM 2.5 exposure [1,[20][21][22][23]. Of these four diseases, myocardial infarction, stroke, and COPD were further sorted by progression period (e.g., first year and following years) because transition probabilities, quality of life, and treatment cost vary significantly [24][25][26]. To investigate the long-term effect of PM 2.5 exposure, the analysis period was 60 years, and the cycle length was 1 year. The outcomes were presented as US dollars (USD) and QALYs, and the costs were discounted at 5% annually to reflect people's positive time preference [27,28]. TreeAge Pro ® 2020 software was used to build the simulation model.

Target Population
A hypothetical cohort of 10,000 Korean women aged 40 years old was analyzed. We targeted middle-aged women because many previous studies on PM 2.5 have been performed targeting middle-aged female population [29,30]. To assess the cost and health risk of increased exposure of PM 2.5 , the study population was assumed to be consistently exposed to either increased or standard concentration of PM 2.5 . Increased concentration was defined as 25 µg/m 3 , which is the average annual PM 2.5 concentration of South Korea in 2017 [6]. Standard concentration was set to 15 µg/m 3 , which is the interim target 3 (IT-3) concentration of PM 2.5 established by WHO.

Input Data
A systematic review was performed to obtain the increased risk of disease incidence due to PM 2.5 exposure (Supplementary Materials) [31]. Firstly, the search terms and PICO (population, intervention, comparison, outcomes) were set, and then we searched through PubMed and conducted an additional search in Google Scholar. The search strategy is shown in Table 1. After the search of electronic database, detailed criteria such as the age of the population or exposure concentration to PM 2.5 were checked by two authors (G.C, Y.K) Any disagreement between the two authors over the eligibility of studies was resolved through discussion with a third author (G.S). Data extraction included sample size of study, age, gender, PM 2.5 exposure status, PM 2.5 concentration increment, and outcomes (relative risk). Cohort studies with large sample size were preferred and final selection was based on the similarity of the study cohort to our target population. Only direct medical costs were included, and non-medical costs such as transportation cost or lost productivity cost were excluded. Domestic studies were preferentially searched since the treatment costs vary by country. Each cost was adjusted by the medical care component of the Consumer Price Index (CPI) in Korea using the equation below [32]. The adjusted costs were then transferred to 2020 US dollars [33].
Costs Current year = Costs Base year × CPI Current year CPI Base year QALY was chosen as a tool to quantify the impact of PM 2.5 exposure on health-related quality of life. The utilities of respective health states were obtained through literature search. The baseline utilities of age and sex-specific Korean general population were sourced from Korea National Health and Nutrition Examination Survey [34]. The utilities of event-free women of standard and increased exposure group were assumed to be the same, which is a conservative assumption.

Sensitivity Analysis
Univariate and probabilistic sensitivity analyses were performed to investigate the robustness of the model because our study was based on several assumptions. Univariate sensitivity analysis was conducted on discount rate (0%, 3%, 7%), time horizon (5, 10, 20, 40 years), and relative risks (95% confidence interval). For the probabilistic sensitivity analysis (PSA), 10,000 times of second-order Monte Carlo simulations were conducted on the relative risks, utilities, and costs. We applied a lognormal distribution for relative risks, a beta distribution for utilities, and a gamma distribution for costs, with the reference of previous studies. The applied distribution for each variable is presented in Table 2. The PSA result was visualized by a scatterplot.

Input Data
For the relative risks, eight studies were selected through systematic review [29,30,[35][36][37][38][39][40]. Relative risk data from the eight studies applied in our model and each reference are shown in Table 3. Among the eight studies, two of them were meta-analysis studies and six were cohort studies. The study populations were from the US, Canada, Europe, South America, and Taiwan. The sizes of the study cohorts were from 65,893 to 367,383 and the follow-up period was from 6 to 14 years. The incremental PM 2.5 concentration was 10 µg/m 3 in five studies and 5 µg/m 3 in three studies. Because the data that matched the characteristics of our target population were not available from domestic studies, relative risks were sourced from international studies. Incidence and mortality rates for each disease were sourced from Korea Statistics. The annual incidence rates, mortality rates, and relative risks used in the model are summarized in Table 3. Table 3. Annual incidence and mortality rates for each disease states and the relative risks related to PM 2.5 exposure used in the model.  Annual costs and QALY data applied for the model and their references are summarized in Table 4. The costs for each health states were referred from domestic studies including cost data estimated from the Korean National Health Insurance database. The age and sex-specific EQ-5D of the general Korean population was sourced from Korea National Health and Nutrition examination survey (2015) [34].

Base-Case Analysis
The model estimated that the increased exposure to PM 2.5 would cost USD 9309 per woman for lifetime healthcare, whereas the lifetime healthcare cost would be USD 8367 per woman when exposed to the standard PM 2.5 concentration. The predicted QALYs were 32.08 and 30.64 for increased exposure and standard exposure, respectively ( Figure 2). The lifetime healthcare cost increased by 11.3% and QALYs decreased by 4.5% in the case of increased exposure to PM 2.5 . 2). The lifetime healthcare cost increased by 11.3% and QALYs decreased by 4.5% in the case of increased exposure to PM2.5.

One-Way Sensitivity Analysis
The results of one-way sensitivity analysis are summarized in Table 5. The one-way sensitivity analysis demonstrated that exposure to an increased concentration of PM2.5 generally shows higher healthcare costs and lower QALYs compared to the standard exposure group across various assumptions. Specifically, when the relative risk of lung cancer incidence was varied, the negative impact of the increased exposure to PM2.5 was the highest resulting in a 23% increase in lifetime healthcare cost. When the discount rate was changed by 0%, 3%, and 7%, the costs were USD 38,589, USD 15,753, and USD 5800 at the increased exposure, which were 7.2%, 9.8%, and 12.6% increases compared to the cost of standard exposure, respectively. As the time horizon increased, the direct healthcare costs of PM2.5 exposure escalated from USD 189 for 5 years to USD 8254 for 40 years. This indicates that the negative economic impact associated with PM2.5 exposure increased over time, and Figure 2 also suggests that the negative impact accelerates as time progresses. Table 5. One-way sensitivity analyses for Korean adult women who are exposed to increased concentration of ambient PM2.5 compared with women exposed to standard concentration.

One-Way Sensitivity Analysis
The results of one-way sensitivity analysis are summarized in Table 5. The one-way sensitivity analysis demonstrated that exposure to an increased concentration of PM 2.5 generally shows higher healthcare costs and lower QALYs compared to the standard exposure group across various assumptions. Specifically, when the relative risk of lung cancer incidence was varied, the negative impact of the increased exposure to PM 2.5 was the highest resulting in a 23% increase in lifetime healthcare cost. When the discount rate was changed by 0%, 3%, and 7%, the costs were USD 38,589, USD 15,753, and USD 5800 at the increased exposure, which were 7.2%, 9.8%, and 12.6% increases compared to the cost of standard exposure, respectively. As the time horizon increased, the direct healthcare costs of PM 2.5 exposure escalated from USD 189 for 5 years to USD 8254 for 40 years. This indicates that the negative economic impact associated with PM 2.5 exposure increased over time, and Figure 2 also suggests that the negative impact accelerates as time progresses. Table 5. One-way sensitivity analyses for Korean adult women who are exposed to increased concentration of ambient PM2.5 compared with women exposed to standard concentration.

Probabilistic Sensitivity Analysis
Probabilistic sensitivity analysis showed that the costs of the increased concentration group varied from USD 5570 to USD 14,045 (134.6%), while the costs of the standard concentration group varied from USD 5382 to USD 12,628 (152.1%). However, the variation for QALYs was smaller (81.7% vs. 88.4%) between the two groups ( Table 6). The result of the probabilistic sensitivity analysis is visualized in Figure 3.

Model Validation
External validation was performed to compare our result to actual observed epidemiological mortality data. In our study, we compared the lung cancer mortality projection result with the observed data reported by Li et al. [21]. The study involved a cohort of 118,551 final participants, 58.9% of which were women, and the follow-up period was 15 years. The cohort was exposed to 31-54 μg/m 3 of PM2.5. For direct comparison, the analysis period of our Markov model was set to 15 cycles. Li and colleagues reported 77.34 lung cancer deaths per 100,000 persons per year for a PM2.5 exposed condition, which was higher than our study projection (56.07 lung cancer deaths per 100,000 persons per year (Table 7)) yet understandable, given the difference in the PM2.5 exposure.

Model Validation
External validation was performed to compare our result to actual observed epidemiological mortality data. In our study, we compared the lung cancer mortality projection result with the observed data reported by Li et al. [21]. The study involved a cohort of 118,551 final participants, 58.9% of which were women, and the follow-up period was 15 years. The cohort was exposed to 31-54 µg/m 3 of PM 2.5 . For direct comparison, the analysis period of our Markov model was set to 15 cycles. Li and colleagues reported 77.34 lung cancer deaths per 100,000 persons per year for a PM 2.5 exposed condition, which was higher than our study projection (56.07 lung cancer deaths per 100,000 persons per year (Table 7)) yet understandable, given the difference in the PM 2.5 exposure. Table 7. Result of the model validation analysis.

Discussion
Our model estimated that adult Korean women exposed to an increased concentration of PM 2.5 incurred an additional USD 942 in their lifetime and lived 1.44 QALYs shorter compared to the standard exposure group. The one-way sensitivity analysis showed that higher healthcare costs and shorter QALYs were expected for the increased exposure group, regardless of various assumptions. For model validation, the results were compared with the external literature, which studied the relative risk of lung cancer death due to PM 2.5 exposure. The predicted mortality rate from our model was 0.005607, which was slightly lower than the observed mortality rate of 0.007734 reported by Li et al. (2020) [21]. However, this difference can be explained by the fact that the mortality rate due to lung cancer in China is higher than in South Korea [49], and the study population of Li et al.
(2020) was exposed to 31-54 µg/m 3 of PM 2.5, which is higher than 25 µg/m 3 , the exposure concentration of the hypothetical cohort in our model. For further validation, the incidence rate of stroke was compared to that of another study [50] where the observed incidence rate was 393 cases per 100,000 person years (0.03930). Our estimation (0.03011) was comparable to the value obtained from the observation data. Therefore, we concluded that our model is valid, and the result of our model is acceptable. Though the data are not shown, we estimated life year expectancy of each cohort by rewarding each cycle being 1 without adjusting the quality of life. When the disease burden was not considered, the expected life years of 40-year-old Korean women were 39.47 years and 41.34 years for increased exposure and standard exposure, respectively. The life expectancy of 40-year-old Korean women was 47.3 years in 2020, based on the lifetable reported by Statistics Korea [51]. This shows that our study provides a conservative estimate.
The economic loss and health impact due to ambient particulate matters has been reported in several previous studies, yet our study is the first attempt to project the longterm effect by using a simulation model. While most studies reported the PM 2.5 -induced economic loss as a regional unit [12], Yin et al. (2017) reported that the PM 2.5 concentration in Beijing (40.26-92.30 µg/m 3 ) induces an economic loss of USD 18 to 147 per capita yearly [52]. These data were calculated by the Willingness to Pay (WTP) or Amended Human Capital (AHC) method, and they include the disutility of illness, productivity loss, medical expenditures associated with illnesses, and expenditures on disease prevention. Because the projection method used in our study and Yin et al.'s (2017) study is different in nature, it is not appropriate to directly compare the results between the two studies. The relatively low health cost in our study is due to not only the analysis method or exposure concentration difference but also the conservative assumptions defined in our model. In our model, we included four diseases in circulatory, respiratory, and neoplasm (lung cancer), which were known to be highly related to PM 2.5 exposure. However, Yin et al.'s (2017) study included additional endocrine/nutritional/metabolic diseases, mental and behavioral disorders, and nervous system diseases and this could increase the cost.
The Markov model method was used in our study to extrapolate the lifetime effect of PM 2.5 based on the data adopted from the existing literature, such as transition probabilities between health states. However, because of this, there are some methodological limitations in our study. First, the relative risks for the diseases used in our model were derived from international studies. The systematic review by Lim et al. (2020) reported the hazard ratios for mortalities due to PM 2.5 increase in the Korean population [53]. However, this study was conducted targeting only the elderly population and could not represent the mortality of middle-aged Korean women. Kim et al. (2018) studied 570 thousand deaths across three metropolitan cities in Korea and reported that PM 2.5 is significantly associated with daily mortality of all causes, and respiratory and cardiovascular diseases [54]. According to the study, the estimation can be updated when the relative risks of the domestic population suitable for our model is reported. Secondly, the effect of PM 2.5 on the economic cost and health outcomes may have been underestimated since only four diseases were included in the model. There is gaining evidence in the relationship between fine particulate matter exposure and various diseases. Some studies have reported that exposure to PM 2.5 is related to the increased morbidity of asthma attacks, diabetes, obesity, Alzheimer's, Parkinson's, dementia, mild cognitive disorders, and bladder cancer [52,55], yet the clinical relevance is inconsistent [1] and further study is needed. In addition, we assumed that the relative risks were constant regardless of the exposure period, which is a clear limitation. However, the effect of PM 2.5 is likely to accumulate for prolonged exposure [56,57]. Finally, although we focused on the effect of PM 2.5 in this study, there are various environmental factors that we did not put into the model, such as toxic elements and possible medicinal interference during the cycles, since the quantified impact (such as relative risks) was either not available or not statistically significant [58,59]. Despite those limitations, our study is the first attempt to project the economic and quality of life impact of PM 2.5 exposure based on a simulation model, which could eliminate the effect of variables other than the exposure to PM2.5 itself.

Conclusions
The negative impact of PM 2.5 was higher on the healthcare costs than on the QALYs, and accelerated as the exposure time accumulated. The results were consistent across various assumptions. A prompt, aggressive intervention is needed to reduce burdens associated with PM 2.5 exposure.