Diseases Burden of Chronic Obstructive Pulmonary Disease (COPD) Attributable to Ground-Level Ozone in Thailand: Estimates Based on Surface Monitoring Measurements Data

Background: Ambient ozone (O3) pollution has increased globally since preindustrial times. At present, O3 is one of the major air pollution concerns in Thailand, and is associated with health impacts such as chronic obstructive pulmonary disease (COPD). The objective of our study is to estimate the burden of disease attributed to O3 in 2009 in Thailand based on empirical evidence. Methods: We estimated disability-adjusted life years (DALYs) attributable to O3 using the comparative risk assessment framework in the Global Burden of Diseases (GBD) study. We quantified the population attributable fraction (PAF), integrated from Geographic Information Systems (GIS)-based spatial interpolation, the population distribution of exposure, and the exposure-response coefficient to spatially characterize exposure to ambient O3 pollution on a national scale. Exposure distribution was derived from GIS-based spatial interpolation O3 exposure model using Pollution Control Department Thailand (PCD) surface air pollution monitor network sources. Relative risk (RR) and population attributable fraction (PAF) were determined using health impact function estimates for O3. Result: PAF (%) of COPD attributable to O3 were determined by region: at approximately, Northern = 2.1, Northeastern = 7.1, Central = 9.6, Eastern = 1.75, Western = 1.47 and Southern = 1.74. The total COPD burden attributable to O3 for Thailand in 2009 was 61,577 DALYs. Approximately 0.6% of the total DALYs in Thailand is male: 48,480 DALYs; and female: 13,097 DALYs. Conclusion: This study provides the first empirical evidence on the health burden (DALYs) attributable to O3 pollution in Thailand. Varying across regions, the disease burden attributable to O3 was 0.6% of the total national burden in 2009. Better empirical data on local specific sites, e.g. urban and rural areas, alternative exposure assessment, e.g. land use regression (LUR), and a local concentration-response coefficient are required for future studies in Thailand.

dramatically exceeded the standard level in many areas of Thailand because of rapidly expanding cities, increasing population density, increasing trend in the number of fossil fuel vehicles and electricity generation (Ruchirawat, Settachan, Navasumrit, Tuntawiroon, & Autrup, 2007;Thanh & Lefevre, 2000). O 3 can be formed in mega cities and carried toward rural areas and distributed across the country via atmospheric transport pathways, therefore, it is necessary to quantify the health burden of O 3 pollution at the national level.
Global estimates of ozone by the powerful chemistry transport model may be under or over-estimating the results because it does not use national level empirical data at the local level. Thailand has used O 3 surface monitored measurements from the Pollution Control Department in recent years, which when used with data from geographic information system (GIS), can improve simulated O 3 distribution at a specific level. GIS is a well-known program and has been used to estimate the exposure distribution in many environmental epidemiology studies including O 3 pollution exposure (Moral García et al., 2010;Hunova, 2011;Nuckols & Lars Jarup, 2004;Verónica, 2013). To estimate the health burden attributed to O 3 in Thailand, it is necessary to develop an accurate prediction of the distribution of O 3 exposure values at non-measurement locations with the empirical data available at the local level.
Our main objective was to estimate the attributable burden of disease due to O 3 exposure in Thailand using the spatial interpolation of O 3 concentrations, health impact function and calculated disability-adjusted life years (DALYs), which are a composite metric that measures both deaths and disabilities, combined with the comparative risk assessment method (Lim et al., 2012;Murray, 1994).

Overall Approach to Estimating Burden Attributable to O 3
GBD uses the disability-adjusted life-year (DALY) as developed by the Burden of Disease workgroup at the World Health Organization (WHO) to quantify the burden of disease. DALY is the sum of the years of life lived with a disability (YLD) and years of life lost (YLL). Our methods are based on the GBD 2010 comparative risk assessment (Lim et al., 2012). We use an integrated method that combines exposure assessment based on surface monitoring measurements and comparative risk assessment (CRA) to quantify the burden of disease attributed to ambient O 3 pollution as shown in Table 1. The comparative risk assessment method used to calculate the attributable burden due to O 3 in this study is based on epidemiological evidence for O 3 exposure. Exposure-response relationships for O 3 are derived from epidemiological studies in GBD 2010, to calculate attributable fractions, and are then multiplied by disease burden, and expressed in DALYs attributable to O 3 (WHO, 2004).
Essentially, the estimation of the burden attributable to O 3 consists of three main steps: (1) measuring the total burden of disease associated with the risk factor at the population level, (2) estimating the population attributable fraction (PAF), and finally (3) applying the PAF to the total burden of disease.  Roberts, Voss, & Knight, 2014). Kriging weights the distance between measured locations based on its spatial correlation to produce variograms and a covariance factor for predicting the unknown locations (Wong, Yuan, & Perlin, 2004). In this study, cross-validation analysis was used to evaluate the performance of the spatial interpolation techniques and the uncertainty of the maps (Janssen, Dumont, Fierens, & Mensink, 2008); we used Pearson correlation (r) between the measured values at the monitoring stations and the model predictions. We assessed the model performance by statistical performance indicators comparing the Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE). The equation is given by; Where, X i , X′ i and N are measured, estimated and number of variables.

Relative Risks and Population Attributable Fraction
The population attributable fraction (PAF) has long been used to estimate the proportion reduction of burden that can be attributed to specified risk factors (Levin, 1953;Rockhill, Newman, & Weinberg, 1998) . If these risk factors were eliminated, it can be concluded that the burden would be reduced from these risk factors. Generally, an exposure-based approach to determine the PAF attributed to O 3 requires three components of data: the exposure of the O 3 , the population of exposure (Pe), and the exposure-response relationships or Relative Risk (RR). We calculated the health impact function for O 3 based on a log-linear relationship between relative risk (RR) and concentrations defined by an epidemiological study (M. Jerrett et al., 2009). We used relative risk from this study because it is the first study that showed significant O 3 long-term health impact. In addition, the GBD 2010 only used RR from this study for their estimations. We assumed that the background O 3 concentrations, the relationship between O 3 and health impact in Thailand, are on the same scale as the GBD 2010. The health impact function was evaluated based on the relationship between relative risk (RR) and the change in O 3 concentrations, defined as follows; (4) Where β is the concentration-response coefficient (CR), which is the slope of the log-linear between O 3 concentrations and mortality, and x -x 0 or Δx is the concentration change from baseline conditions (x 0 ), defined as follows: We calculated the burden of disease attributable to ambient O 3 pollution by multiplying the total disease burden and the PAF which was determined by equation 5 (Murray, Ezzati, Lopez, Rodgers, & Vander Hoorn, 2003), where Pe is the population distribution of the exposure and RR is the relative risk. The total COPD disease burden (DALY) dataset was obtained from BOD Thailand (Thai BOD) (BOD, 2009), which published the national disease burden in Thailand (BOD, 2009). We assumed that the ratio of males to females (sex ratio), is on the same scale as the Thai BOD (Bundhamcharoen, Odton, Phulkerd, & Tangcharoensathien, 2011).

Results
We present a number of comparisons of the results using the different spatial modeling described in section 2.
The statistical summary of model performance corresponding to Kriging and IDW interpolation methods are presented in  Figure 3 illustrates the cross validation between the measured values at the monitoring stations and the model predictions. For both models, the Pearson correlation of the IDW model (r = 0.272) was lower than the Kriging model (r = 0.5). In addition, the model performance assessments showed that the MAE and RMSE for O 3 varied from 17.9 to 20.25 and 22.58 to 25.13, respectively. These results revealed that the Kriging interpolation method performed with 10-12% better accuracy than the IDW method, and clearly indicates that the Kriging provides optimal interpolation model for the exposure assessment based on the empirical dataset in the present study. In addition, the spatial distribution of O 3 concentration from the IDW and Kriging model are presented in Figure 4. The Kriging model clearly indicates that the highest O 3 concentration levels in the Central and Eastern regions with a range of 117-131 ppb and, the lowest levels in southern region, with a range of 69-83 ppb.
We then calculated ΔX in six regions and discovered that the average ΔX values were between 64.7-84.9 ppb in Thailand (Table 4). We calculated the summation of PAF grids ( Figure 4) by region, as shown in Table 4. Our best PAF estimation indicates that the western region had the lowest PAF value, or about 1.47% of the total COPD burden. The highest PAF value was located in the central region, which produced 9.6% of the total COPD burden reflecting the highest O 3 concentration levels.

Discussion
This study estimated the health burden of diseases due to O 3 , using GIS interpolation models and health impact function, and followed the comparative risk assessment method (CRA) from the global burden of disease (GBD) study. This approach makes it possible to assess the burden attributable to empirical data sources for O 3 exposure, e.g., surface monitor measurements, while taking into consideration the population exposure at the national level.
Our results indicate that the burden of disease attributable to O 3 is approximately 61,577 DALYs (0.6% of total DALYs) (male: 48,480 DALYs; female: 13,097 DALYs). The only source of relative risks due to O 3 from the GBD study is Jerrett et al. (2009), a major cohort study in the United States and also the first study to show significant O 3 exposure long-term health impacts.
We estimated O 3 exposure by two spatial exposure-modeling techniques, i.e., IDW and Kriging. The cross-validation technique was used to test the model accuracy from the air pollution exposure study (Thepanondh & Toruksa, 2011), because it does not require additional data collection from areas without an air surface monitoring station. It removes a known point from the interpolated dataset and the residual dataset is used to evaluate the prediction at the known point that it had been removed. The model performance indicator of each of the interpolation methods is shown in Table 4, and the predicted cross-validation data is shown in Figure  2. Both interpolation techniques were useful for estimating the exposure from the unknown areas, but the results showed that the Kriging, which produced smaller RMSE and MAE values, performed better than IDW. Using the Kriging method as shown in Figure 4, an O 3 concentration exposure map from spatial distributed interpolation was produced.
From our study, the average ΔX value in Thailand = 69.3 ppb (95% CI 62-70). These figures are larger than Horowitz's simulation at 10-25 ppb (Horowitz, 2006); Horowitz used the chemistry transport model to generate globally simulated O 3 concentrations of 30-50 ppb in area of Thailand. To improve these results at the national level, increasing the number and distribution of monitoring stations is required in the future because of a lack of representative surface monitored data in some high population provinces (populations > 1 million) in Thailand. www.ccsenet.org/gjhs Global Journal of Health Science Vol. 8, No. 1; For example, Ubon ratchathani, Udon Thani and Buriram in northeastern region, Nakhon Si Thammarat in southern region, all high population provinces, lack representative surface monitored data. Figure 4 illustrates how the population attributable fraction due to O 3 exposure varies across Thailand. The population attributable fraction was the highest in the Central region (including Bangkok Metropolitan Area), most likely because the higher population distribution of exposure (figure 2) and O 3 concentrations levels (figure 4) are located in this region. Furthermore, we found that the spatial variation of population attributable fractions were generally more consistent with the population distribution of exposure than the O 3 concentrations, similar to the previous study that suggested the exposure comes with population; no population, no exposure (Hao, Flowers, Monti, & Qualters, 2012). The highest attributable COPD burden due to O 3 was located in Central region, or about 24,812 DALYs. This may be due to the population distribution of exposure and pollution levels, but also because the total COPD DALY in the Central region was consistently high as well. However, the Northern region, which had the highest COPD DALY in Thailand, had lower estimated O 3 attributable disease burden than the Central region (5,490 DALYs). This may be because the O 3 concentrations were slightly lower (64.7 ppb, 95% CI; 57.9-71.4) than the Central region (84.9 ppb, 95% CI; 81.8-88.04), the population distribution of exposure was relatively smaller than the Central region, and the major sources of air pollution which are associated with COPD in the Northern region may differ from the Central region (e.g. forest fire, and garbage burnings) (Sukitpaneenit & Kim Oanh, 2014;Wiwatanadate, 2014).
In a previous study, GBD 2010 (Lim et al., 2012), estimated the disease burden attributable to O 3 to be approximately 0.1% of the global DALYs (152,434 deaths: both sex) in 2010, our best estimate, however, is 0.45-1%, depending on the province/region, or a total of 61,577 DALYs (0.6% of the total national disease burden). Our results indicate that our PAFs were slightly greater than the GBD study. One explanation may be that the exposure estimation method. GBD used the global chemistry transport models (Fiore, Dentener, Wild, Cuvelier, & Schultz, 2009;Stevenson et al., 2006) to estimate the distribution of O 3 . GBD did not link the available surface monitor data and statistical model to estimate exposure distribution similar to the fine particulate matter (PM2.5) in their study, which reflects a lack of available local empirical data. In addition, the estimations for the O 3 disease burden in Thailand are approximate to the United Arab Emirates (UAE). Based on the air quality monitoring stations in the UAE, and the spatial exposure modeling technique, Ying Li et al. (Li et al., 2010) estimated that disease attributed to ground-level O 3 exposure in UAE was approximately 0.8% (95% CI; 0.2-2%) of the total UAE disease burden in 2007.
We note that our study has some inherent limitations, which need to be addressed. We rely on O 3 pollution data from existing monitoring stations network. In addition, the exposure assessment based on the surface monitoring measurements may vary, depending on location, air quality monitoring stations density and seasonal variation in O 3 pollution which may not be uniformly distributed across Thailand. For that reason, it is possible that our model was overestimates to simulate annual O 3 exposure into large scale area (e.g. national scale). We recommend that future research should consider another ambient air quality model to simulate O 3 concentrations, e.g. land use regression (Ryan & LeMasters, 2007) and photochemical modeling (Fann et al., 2012) to more accurately assess O 3 pollution and population exposure on a national scale.
In the study limitation, we assumed that the exposure was equal for all members of the population ≥ 25 years (AIHW, 2010). For that reason, it should be noted that using an urban/local area-weighting method to estimate population exposure (Zhang, Qi, Jiang, Zhou, & Wang, 2013) due to O 3 , and the higher resolution population dataset at a resolution of 30 arc-seconds (~1km), which allocating census block population and household information into regular latitude longitude grids (Balk et al., 2006), are required to improve the contribution on the Pe factor at the local level.
Another limitation is that our study relies on the health outcome model from the GBD study (Lim et al., 2012), and also concentration-response coefficients (CR) from an epidemiological study (M. Jerrett et al., 2009) despite differences in health status, lifestyle, age structure, and medical care (Anenberg et al., 2010). Therefore, we did not account for short-term effects by the ambient O 3 pollution, which is related to cardiovascular and respiratory mortality (e.g. Bell et al., 2004). These limitations may be eliminated in future research if local epidemiological studies are available.

Conclusion
Our study aims to estimate the disease burden attributed to ozone in Thailand. This study presents an integrated exposure assessment, using a spatial interpolation model from empirical data, population distribution exposure and health impact function to estimate national disease burden attributable to O 3 . Our study indicated that ambient O 3 is one of the major air pollutants that exerts adverse effects on the environment and human health in www.ccsenet.org/gjhs Global Journal of Health Science Vol. 8, No. 1; Thailand. The estimated COPD burden cause by O 3 in our study was about 0.6% of the national burden each year. It is our hope that the results will initiate more precise O 3 and/or other air pollution health burden estimations among scientists in future environmental health burden studies. Finally, this study provides the first national estimate and can inform decision-making by stakeholders and policy-makers to promote and manage a health co-benefit of green economy in the future.