A health impact assessment of long-term exposure to particulate air pollution in Thailand

Particulate air pollution causes a spectrum of adverse health effects affecting the respiratory, cardiovascular, neurological, and metabolic systems that are hypothesised to be driven by inflammation and oxidative stress. Millions of premature deaths each year are attributed to exposure to ambient particulate matter (PM). We quantified health and economic impacts from long-term exposure to ambient PM2.5 in the population of Thailand for 2016. We collected data on ambient PM2.5 concentrations from automatic monitoring stations across Thailand over 1996–2016. We used historic exposure to PM2.5 to estimate the mortality in each province from lower respiratory infections (LRIs), stroke, chronic obstructive pulmonary disease, lung cancer, and ischaemic heart disease, and also assessed diabetes mortality, as well as incident cases of dementia and Parkinson’s disease, in supplementary analyses. We applied risk estimates from the Global Exposure Mortality Model to calculate attributable mortality and quantify disability-adjusted life years (DALYs); we based economic costs on the value of a statistical life (VSL). We calculated 50 019 (95% confidence interval [CI]: 42 189–57 849) deaths and 508 918 (95% CI: 438 345–579 492) DALYs in 2016 attributed to long-term PM2.5 exposure in Thailand. Population attributable fractions ranged from 20% (95% CI: 10% to 29%) for stroke to 48% (95% CI: 27% to 63%) for LRIs. Based on the VSL, we calculated a cost of US$ 60.9 billion (95% CI: US$ 51.3–70.4 billion), which represents nearly 15% of Thailand’s gross domestic product in 2016. While progress has been made to reduce exposure to ambient PM2.5 in Thailand, continued reductions based on stricter regulatory limits for PM2.5 and other air pollutants would help prolong life, and delay, or prevent, onset of cardiorespiratory and other diseases.


Background
Exposure to ambient particulate matter (PM), especially that from traffic-related sources, has consistently been linked to adverse health outcomes (Kim et al 2015). There is now a growing body of evidence suggesting exposure to non-traffic derived PM, including industrial emissions, natural dust and salts, and biomass burning, can also cause harmful health effects (Hime et al 2018, Johnston et al 2019. Air pollution may damage health through a number of mechanisms, including oxidative stress and inflammation, leading to detrimental impacts on the respiratory, and cardiovascular systems, as well as a growing list of other adverse health outcomes (Münzel et al 2017). In addition to PM, gaseous air pollutants, such as ozone, nitrogen dioxide, carbon monoxide, and sulphur dioxide, have been documented to cause a wide range of pathologies (Schraufnagel et al 2019).
The population of Thailand is exposed to PM from a range of sources, which varies spatially and temporally. In the capital, Bangkok, with a population of over 12 million, the number of cars on the road exceeds 10 million, representing a significant source of PM of size <2.5 µm (PM 2.5 ) (Cheewaphongphan et al 2017). Emissions from biomass burning, industry, traffic, and power plants are the main contributors of PM 10 emissions in Thailand (Vongmahadlek et al 2009). In the north, and the rest of the country to a lesser extent, the population is exposed to annual episodes of biomass burning of agricultural residues and forested areas, originating from within and outside Thailand, ultimately leading to very high ambient PM 10 concentrations during January to April each year (Kliengchuay et al 2018, 2021, Punsompong and Chantara 2018, Mueller et al 2020. Health impact assessment (HIA) is a methodology used to estimate the change in health impacts on a population level based on a given change in an exposure. In addition to predicting impacts from different policy options, HIA can also be used to quantify the overall burden of disease from a given exposure. Exposure to ambient PM 2.5 is estimated to lead to over 4 million deaths globally each year (Cohen et al 2017). In Thailand specifically, annual all-cause mortality has been estimated to be almost 40 000, with the percentage of deaths attributed to PM 2.5 at nearly 17% for lung cancer (Pinichka et al 2017).
To update and expand earlier work on the burden of disease from particulate air pollution, we undertook an HIA to quantify the health and economic impacts across Thailand from long-term exposure to ambient PM 2.5 . We included health endpoints where evidence of exposure to ambient particulate air pollution is more established (i.e. respiratory and cardiovascular disease) and also investigated in supplementary analyses those still emerging (i.e. neurological and metabolic diseases). The results of this assessment could be used to provide justification for stricter air pollution regulations, both in Thailand and in other jurisdictions. This study is part of a research project to study the effects of air pollution in Thailand: Thailand Air Pollution Health Impact Assessment (TAPHIA).

Methods
We conducted an HIA, which necessitates identifying pollution sources, population at risk, background health data, and concentration-response functions (CRFs). The data from these inputs are combined in a health impact function to provide estimates of health impacts.

Air pollution data
We obtained from the Thai Pollution Control Department hourly averaged pollutant data from the national ground monitoring network of 63 sites. Data were available on individual pollutants, including PM 10 , PM 2.5 , carbon monoxide (CO), ozone (O 3 ), nitrogen dioxide (NO 2 ), and select volatile organic compounds. Air pollutant data were provided over the period 1996-2017. To measure the long-term exposure to PM 2.5 , we calculated mean concentrations over the period 1996-2016 (2017 was excluded, as health data were received only up to 2016). To ensure that monitors were representative of the full period, we included only background monitors with sufficient data (i.e. those that had <25% missing PM data over 1996(Wong et al 2008). This reduced the number of monitors with sufficient data to 16 sites, which were located in provinces representing 23% of the study population (see figure S1 (available online at stacks.iop.org/ERL/16/055018/mmedia)). We also performed a sensitivity analysis with two additional monitors with less than sufficient data (see supplementary material).
As the measurement of ambient PM 2.5 in Thailand did not start until 2011, we estimated concentrations during 1996-2011 based on the ratio of PM 2.5 :PM 10 from stations where both PM size fractions were collected concurrently. We calculated regional average ratios and then applied them to PM 10 concentrations of stations within each region (WHO 2016a) to calculate surrogate PM 2.5 concentrations. The mean PM 2.5 :PM 10 ratios ranged from 0.50 (south) to 0.78 (northeast), which appear to be similar to published values, such as for Bangkok, Thailand (Fold et al 2020) and China (Gao et al 2015, Xu et al 2016 (see table S1). We calculated means for each station across 1996-2016 using these surrogate and true PM 2.5 concentrations. To convert these point estimates to average values for each province in Thailand, which could then be linked to province-level health and population data in the HIA, we used BenMAP (CE 1.5.0), a mapping software developed by the United States Environmental Protection Agency (USEPA 2018). BenMAP uses Voronoi neighbourhood averaging to interpolate air pollutant concentrations, which has been shown to produce similar results to inverse distance weighting and kriging (Yu et al 2018). As a comparator, we also calculated long-term mean provincial PM 2.5 values based on nearest neighbour analysis.

Population at risk
The population at risk for exposure to air pollution was the whole population of Thailand. However, as we were interested in quantifying the longterm effects of air pollution, we included only those in Thailand aged 25 years or older; mortality levels are quite low in these younger age groups, so their exclusion would not greatly alter our estimates. We obtained population estimates by 5 year age group for each province from the Thai National Statistics Office (2020). For the year 2016, we used a total population figure of 43 831 365.

Selection of health endpoints
We included those causes of mortality with robust evidence with PM 2.5 , namely: lower respiratory infections (LRIs), stroke, chronic obstructive pulmonary disease (COPD), lung cancer, and ischaemic heart disease (IHD) (Burnett et al 2018). In addition to these well-established health outcomes associated with air pollution exposure, we included in the HIA novel endpoints with compelling evidence of harm from more recent epidemiological studies. As important chronic conditions that constitute a growing global burden of disease (Naghavi et al 2017), we included diabetes, Alzheimer's disease and Parkinson's disease in the HIA, which have been linked to long-term PM 2.5 exposure (Dimakakou et al 2018). We present in the main paper the results of the established health endpoints and include in the supplementary material additional details on the novel health outcomes.
It has been reported that Thailand death certificates entail a high proportion of unclassified or nonspecific deaths (Rao et al 2010, Naghavi et al 2017. Therefore, since we obtained cause-specific mortality data, we adjusted upward the rates to account for uncoded deaths. As the GBD study employed a comprehensive methodology to redistribute unclassified deaths for individual countries, we multiplied our calculated provincial-and agespecific rates by a cause-specific mortality correction factor. This adjustment assumes uncoded deaths do not vary by age or province. The correction factors ranged from 1.17 (stroke) to 2.72 (COPD) (see table S2) and were calculated with the following formula: (1)

Concentration response functions
For the better established health endpoints (i.e. LRI, stroke, COPD, lung cancer, and IHD), we used CRFs developed by Burnett et al (2018) in the Global Exposure Mortality Model (GEMM), which is based on 41 cohort studies of exposure to ambient PM concentrations in populations predominantly in Europe and North America, but also in Asia.

Calculation of health impacts
Using the above discussed inputs, we constructed an HIA model in MS Excel to calculate health impacts for the populations of every 5 year age group over 25 years (40+ years for Alzheimer's disease) in each province of Thailand. We used a burden of disease approach, meaning we calculated the cases of mortality/incidence for each health endpoint that was attributed to long-term PM 2.5 exposure, as defined by overall mean concentrations in each province during 1996-2016. We used a counterfactual concentration of 2.4 µg m −3 , which implies there would be no change in risk from 0 to 2.4 µg m −3 (Burnett et al 2018). Therefore, we subtracted 2.4 from the province-level long-term PM 2.5 means for the purposes of health impact modelling. Upper and lower 95% confidence intervals (CIs) were calculated for health and economic impacts based on the CIs of the CRFs; 95% CIs of the aggregate impacts were based on the square root of the sum of squares of the individual standard errors (Altman & Bland 2003). We also modelled the health and economic benefits for two scenarios: universal reductions in mean ambient PM 2.5 concentrations to 10 µg m −3 and 25 µg m −3 in every province, based on the World Health Organization (WHO) and the Thai National Ambient Air Quality Standards annual limits for PM 2.5 , respectively. We calculated the potential differences in impacts by subtracting mortality, incidence, and costs associated with the historic PM 2.5 levels with those calculated in the two scenarios. Actual PM 2.5 levels were retained for any provinces with levels below either of the scenario values; n = 16 provinces had mean PM 2.5 concentrations below 25 µg m −3 .
We used the CRFs from Burnett et al (2018) to calculate deaths from LRI, stroke, COPD, lung cancer, and IHD. Population attributable fractions (PAFs) were calculated based on observed (O) and expected (E) deaths/cases, as in equation (2) (Mansournia and Altman 2018).
We calculated years of lost life (YLL) for each individual cause of mortality by multiplying the number of deaths by the probability of survival in each 5 year age group up to the median life expectancy of 76 years (World Bank 2019) based on a life table for Thailand (Miller andHurley 2003, WHO 2016b). We estimated years living with disability (YLDs) for the incident neurodegenerative diseases and also for mortality from diabetes, IHD, COPD, and lung cancer. YLDs were calculated by applying disability weights to an assumed amount of time living with disability prior to death. Disability weights can take the value of 0 (good health) to 1 (death) depending on the severity and impact of a given disease. We summed YLDs and YLLs to calculate the overall disability adjusted life years (DALYs). We display in table S3 our assumptions for YLD calculations.

Calculation of economic impacts
We estimated the economic costs of the deleterious health impacts attributed to long-term PM 2.5 exposure. This may be calculated using a value of a statistical life (VSL), which can be used to assess tradeoffs between health and safety and costs. VSLs may be based on surveys of willingness-to-pay or wage compensation for riskier jobs. This metric differs from and tends to encompass much higher costs than those specifically incurred by the health care system, which we did not address. We identified a study that estimates a VSL for Thailand using data from 2012 to 2014 (Witvorapong and Komonpaisarn 2020). We adjusted this VSL to account for inflation (i.e. the general rise of prices during that period) to provide a VSL in 2016 of roughly US$ 1.22 million, which is very similar to other estimates of the average VSL for an upper-middle income country (i.e. such as Thailand) (Viscusi and Masterman 2017). As a comparator, we adjusted the 2005 Organisation for Economic Co-operation and Development (OECD) VSL of US$ 3 million based on purchasing power parity and using an income elasticity of 0.9 (OECD 2016) to calculate an amount of US$ 1.65 million; this adjustment was necessary to account for the cost of living in Thailand. We multiplied each of these VSL figures by the number of deaths attributed to PM exposure.
We estimated the value of a statistical life year (VSLY), which accounts for the projected YLLs, not just mortality. This such measure is considered contentious, as it implies a lower value assigned to deaths at older ages, though we included this metric to attempt to quantify the burden of premature deaths; the VSL may overstate costs, since ultimately deaths are delayed, rather than prevented (Martenies et al 2015). We calculated VSLY by dividing the VSL by the life expectancy in Thailand (76 years). This calculation provided an average cost for a life year, assuming no difference in the value given to different ages, of US$ 16 008 and US$ 21 682, based on the Thai and OECD adjusted VSLs, respectively. To help interpret results, we compared the overall economic costs as a percentage to the gross domestic product (GDP) of Thailand in 2016, which was US$ 412 billion (Country Economy 2020).

Results
The estimated long-term (1996-2016) mean PM 2.5 concentrations for each province in Thailand ranged from 20.5 µg m −3 to 37.4 µg m −3 . Ambient concentrations were lowest in the south of the country, and highest in Bangkok, as well as areas in the north (see figure 1). There was an apparent longterm decline in PM 2.5 levels across the monitoring network, with mean annual concentrations decreasing from 41.5 µg m −3 in 1996 to 24.1 µg m −3 in 2016 (−41.9%). The reduction in ambient concentrations appears mostly to have occurred in the first half of the study period, with levels mainly constant in the past decade (see figure 2). Although there was good correlation between the Voronoi averaging and those values calculated using nearest neighbour (r = 0.76), concentrations were generally greater using the latter method (data not shown).

Discussion
We undertook an HIA for long-term exposure  to PM 2.5 for the population of Thailand aged over 25 years in 2016. We calculated 50 019 deaths and over 508 918 DALYs from LRI, stroke, COPD, lung cancer, and IHD. In modelling a nation-wide reduction to 10 µg m −3 of PM 2.5 , we identified decreases in the overall PAF of 18 percentage points.
Our HIA in Thailand from air pollution mortality estimates were somewhat smaller than those calculated for the year 2015 by Burnett et al (2018) using the GEMM estimates, ranging from 1% (COPD) to 39% (LRI) lower. These differences can largely be explained by disparities in the baseline mortality data. We consolidated individual deaths by province and corrected for uncoded deaths using national rates from the GBD study (Naghavi et al 2017), whereas the GEMM study made use of the Global Health Data Exchange, which combines numerous different global data sources; to illustrate the differences, the baseline LRI mortality counts was 72% higher in the latter study. We believe our study provides reliable provincial baseline mortality rates for 2016. Pinichka et al (2017) calculated 38 410 deaths for all-causes in Thailand in those over 30 years of age in 2009. Our estimates, based on specific causes of diseases are 30% in excess of this figure, likely due to a number of factors, including the application of greater estimates of risk (i.e. GEMM) across larger overall reductions in PM 2.5 . The PAFs calculated from exposure to fine PM in our study were substantial, with values for individual causes of mortality ranging from 20% to 48%. For those endpoints for which we relied on GEMM risk estimates, our PAF values are in line with those projected for the observed magnitude of exposure levels (Burnett and Cohen 2020).
We modelled two hypothetical scenarios in which there were consistent PM 2.5 reductions in each province to 25 and 10 µg m −3 . Although these might be challenging reductions to achieve, particularly in the higher PM 2.5 settings, for example, in Bangkok, a study reported a decrease of 21.8% in mean PM 2.5 concentrations in an urban area in Thailand during March 2020 due to COVID-19 mitigation measures, compared to the preceding time period (Stratoulias and Nuthammachot 2020); other such studies in Southeast Asia also have found reductions in PM 2.5 and other air pollutants (Kanniah et al 2020). It is encouraging that such reductions are possible even though the measures needed to bring them about have been drastic; it remains to be seen whether these reductions are temporary, or whether there will be a priority on sustaining them as societies get accustomed to, and overcome, COVID-19. While even more modest reductions of ambient PM 2.5 levels will be beneficial, any potential decreases in attributable deaths in the future may be offset to some extent by a growing and ageing population (Butt et al 2017).
HIAs in other jurisdictions have estimated economic costs of ill health attributed to air pollution. Due to the wide range of valuation that can be used in an HIA, particularly given the economic disparities between jurisdictions, it is difficult to compare monetary impacts on health. Trejo-Gonzalez et al (2019) estimated the number of deaths and costs that would have been prevented if PM 2.5 levels in Mexico were reduced to 10 µg m −3 , which equated to 14 666 deaths (all-cause) and US$ 64 billion, based on the OECD VSL (US$ 3 million) and adjusted for purchasing power in Mexico. In Japan, Seposo et al (2018) calculated nearly 25 000 attributable deaths (all-cause), at a cost of nearly US$ 80 billion with a VSL of US$ 3.2 million. These cost estimates from particulate air pollution are within the range of those calculated in the present study using the Thai and adjusted OECD VSL (i.e. US$ 60.9-82.4 billion).
We included PM 2.5 as the main exposure in this HIA, since research has indicated this pollutant entails the most widespread health impacts (Cohen et al 2017). Studies demonstrate that other air pollutants have also been linked to ill health, but often air pollutants are correlated and are believed to act on similar pathways (e.g. PM 2.5 and NO 2 ), so including multiple pollutants in an HIA would likely overestimate health impacts. Using unadjusted CRFs for PM 2.5 in HIAs also accounts to some extent for health impacts caused by other pollutants from common sources (COMEAP 2018). There were important sources of uncertainty with the various inputs to our calculations, which is typical in HIA studies. Previous work indicates much of the uncertainty with mortality estimates may be due to the PM 2.5 estimates and the CRFs, rather than baseline mortality rates (Giani et al 2020), despite the differences in the latter between ours and other studies. We estimated mortality from exposure to PM 2.5 based on mean concentrations of long-term periods (21 years). This is a longer exposure period than that included in other HIAs, which have based mortality from 1 (Pinichka et al 2017) to 5 (Cohen et al 2017) years. The GEMM risk estimates are based on cohort studies of multi-year follow-up, meaning a multi-year exposure period would be most appropriate, though the optimal period length is unclear. Given the relatively lower recent ambient concentrations in Thailand, had we based exposures on a shorter and more recent period, attributable mortality in general would have been reduced, depending on the specific period and duration. By contrast, if we had explicitly accounted for those outcomes with longer latencies or preclinical periods (e.g. lung cancer), such as more heavily weighting or only using the more historic (higher) exposure period, the calculated health impacts would have been greater. Quantification of mortality in a single year based on multiple years of exposure would not capture deaths from earlier years and thus would underestimate the true impact. Likewise, long-term annual averages would to some degree mask seasonal peaks in ambient concentrations, for which there would likely be short-term increases in mortality, for example, the seasonal biomass burning in northern Thailand (Pothirat et al 2019). PM 2.5 estimates based on nearest neighbour methods were found to be higher in our study, although such methods have been shown previously to overestimate concentrations (Xie et al 2017). In addition, as we only included the subset of population older than 25 years, our mortality figures are necessarily an underestimate of the impacts from air pollution, though the contribution from younger ages is small by comparison (Lelieveld et al 2018).
Our study raises a number of important considerations to be addressed in future research and HIA studies. We included a broader suite of health endpoints in the supplementary analyses, which produced important findings, as the second highest DALYs were from Alzheimer's disease and dementia. As the evidence base becomes more firm and precise, these and other health outcomes should be considered for quantifying the overall burden of disease and estimating impacts related to ambient air pollution control policies. The duration of exposure data will depend on the specific period of interest, but should be aligned with risk estimates from research with a comparable study period. HIAs should be undertaken to assess health and economic impacts from short and long-term changes as a result of policy measures and large-scale behaviour changes possibly related to COVID-19.

Strengths and limitations
We attempted to use the best available data for all inputs to provide accurate estimates of the potential health impacts in Thailand from long-term exposure to PM 2.5 . Unlike previous HIAs, we included provincial level exposure and corrected health data to help refine our estimates, as well as incorporated recent evidence on novel health endpoints. In addition, we extended mortality impacts to include DALYs and economic impacts, and we examined effects of PM 2.5 reductions in two scenarios based on national regulatory and WHO limits. The extended period of air pollutant monitoring data (21 years) that we included would have more represented cumulative lifetime exposure. Nevertheless, there were limitations in our calculations. We acknowledge that the network of air quality monitors is not completely spatially representative of Thailand. However, the longterm averages between the different sites were within 20 µg m −3, so it is not anticipated that using data from additional monitors in less populated (i.e. more rural) areas would have a significant effect on the calculated population level health impacts. Still, interpretation of our results likely will be more reliable at the national level, as potential differences in PM concentrations and sources may impact findings for a given province. We did not have PM 2.5 concentrations for the entire exposure period and had to extrapolate using levels of PM 10 . We excluded effects from concentrations below 2.4 µg m −3 , for which concentration response curves are not known, but may also entail adverse health. We corrected as much as the data would allow for uncoded mortality and adjusted for age-related rates; the correction factors we applied were based on national averages, which may have attenuated any differences between provinces. We did not account for mortality from exposure to ambient ozone, which is often less correlated with PM 2.5 ; however, less than 10% of global deaths from ambient air pollution have been attributed to ozone, so this exclusion is not likely to substantially increase our estimates (Cohen et al 2017). Further, a meta-analysis found little evidence of increased mortality with longterm exposure to ozone (Atkinson et al 2016).

Conclusions
We have completed an HIA indicating that longterm exposure to PM 2.5 has resulted in approximately 50 000 annual deaths in Thailand. Overall costs of this excess mortality is equivalent to nearly 15% of Thailand's GDP in 2016, based on the value of a statistical life. Our study benefitted from a lengthy longitudinal dataset of ambient air quality, mostly provincespecific health data, and established and emerging risk estimates of mortality and incidence. While progress has been made to reduce exposure to ambient PM 2.5 in Thailand, continued reductions based on stricter regulatory limits for PM 2.5 and other air pollutants would help prolong life, and delay, or prevent, onset of cardiorespiratory and other diseases.

Data availability statement
The data that support the findings of this study are available upon reasonable request from the authors.