key: cord-0257836-67rsh5ob
authors: Yin, H.; Brauer, M.; Zhang, J.; Cai, W.; Navrud, S.; Burnett, R.; Howard, C.; Deng, Z.; Kammen, D. M.; Schellnhuber, H. J.; Chen, K.; Kan, H.; Chen, Z.; Chen, B.; Zhang, N.; Mi, Z.; Coffman, D.; Wei, Y.; Cohen, A.; Guan, D.; Zhang, Q.; Gong, P.; Liu, Z.
title: Global Economic Cost of Deaths Attributable to Ambient Air Pollution: Disproportionate Burden on the Ageing Population
date: 2020-05-05
journal: nan
DOI: 10.1101/2020.04.28.20083576
sha: 27f2f03b9712072811d154646c745fd6940be951
doc_id: 257836
cord_uid: 67rsh5ob

Summary Background The health impacts of ambient air pollution impose large costs on society. While all people are exposed to air pollution, older individuals tend to be disproportionally affected. As a result, there is growing concern about the public health impacts of air pollution as many countries undergo rapid population ageing. We investigated the spatial and temporal variation in the health economic cost of deaths attributable to ambient air pollution, and its interaction with population ageing from 2000 to 2016 at global and regional levels. Methods We developed an age-adjusted measure of the value of a statistical life year (VSLY) to estimate the health economic cost attributable to ambient PM2.5 pollution using the Global Burden of Disease 2017 data and country-level socioeconomic information. First, we estimated the global age- and cause-specific mortality and years of life lost (YLL) attributable to PM2.5 pollution using the global exposure mortality model (GEMM) and global estimates of exposure derived from ground monitoring, satellite retrievals and chemical transport model simulations at 0.1 by 0.1 degree (~11 km at the equator) resolution. Second, for each year between 2000 and 2016, we translated the YLL within each age-group into a health-related economic cost using a country-specific, age-adjusted measure of VSLY. Third, we decomposed the major driving factors that contributed to the temporal change in health costs related to PM2.5. Finally, we conducted a sensitivity test to analyze the variability of the estimated health costs to four alternative valuation measures. We identified the uncertainty intervals (UIs) from 1000 draws of the parameters and exposure-response functions by age, cause, country and year. All economic values are reported in 2011 purchasing-power-parity-adjusted US dollars. Findings Globally, 8.42 million (95% UI: 6.50, 10.52) deaths and 163.68 million (116.03, 219.44) YLL were attributable to ambient PM2.5 in 2016. The average attributable mortality for the older population was 12 times higher than for those younger than 60 years old. In 2016, the global health economic cost of ambient PM2.5 pollution for the older population was US$2.40 trillion (1.89, 2.93) accounting for 59% of the cost for the total population. The health cost for the older population alone was equivalent to 2.1% (1.7%, 2.6%) of global gross domestic product (GDP) in 2016. While the economic cost per capita for the older population was US$2739 (2160, 3345) in 2016, the cost per capita for the younger population was only US$268 (205, 335). From 2000 to 2016, the annual global health economic cost for the total population increased from US$2.37 trillion (1.88, 2.87) to US$4.09 trillion (3.19, 5.05). Decomposing the factors that contributed to the rise in health economic costs, we found that increases in GDP per capita, population ageing, population growth, age-specific mortality reduction, and PM2.5 exposure changed the total health economic cost by 77%, 21.2%, 15.6%, -41.1% and -0.2%, respectively. Compared to using an age-invariant VSLY or an age-invariant value of a statistical life (VSL), the estimates of the older population share of the total health economic cost using an age-adjusted VSLY was 2 and 18 percentage points lower, respectively. Interpretation The health economic cost borne by the older population almost doubled between 2000 and 2016, driven primarily by GDP growth, population ageing and population growth. Compared to younger individuals, air pollution leads to disproportionately higher health costs amongst the older population, even after accounting for their relatively shorter remaining life expectancy and increased disability. The age-specific estimates of health economic cost inform the optimal design of air pollution reduction strategies and allocation of healthcare resources. The positive relationship between age and economic costs suggests that countries with severe air pollution and rapid aging rates would particularly benefit from improving their air quality. In addition, strategies aimed at enhancing healthcare services, especially for the older population, may be beneficial for reducing the costs of ambient air pollution.

Background 40

The health impacts of ambient air pollution impose large costs on society. While all people are 41 exposed to air pollution, older individuals tend to be disproportionally affected. As a result, 42

there is growing concern about the public health impacts of air pollution as many countries 43 undergo rapid population ageing. We investigated the spatial and temporal variation in the 44 health economic cost of deaths attributable to ambient air pollution, and its interaction with 45 population ageing from 2000 to 2016 at global and regional levels. 46 47 Methods 48 We developed an age-adjusted measure of the value of a statistical life year (VSLY) to estimate 49 the health economic cost attributable to ambient PM2.5 pollution using the Global Burden of 50

Disease 2017 data and country-level socioeconomic information. First, we estimated the global 51

age-and cause-specific mortality and years of life lost (YLL) attributable to PM2.5 pollution 52

using the global exposure mortality model (GEMM) and global estimates of exposure derived 53 from ground monitoring, satellite retrievals and chemical transport model simulations at 0.1 54 × 0.1 (~11 km at the equator) resolution. Second, for each year between 2000 and 2016, we 55 translated the YLL within each age-group into a health-related economic cost using a country- 56 specific, age-adjusted measure of VSLY. Third, we decomposed the major driving factors that 57

contributed to the temporal change in health costs related to PM2. 5 factors that contributed to the rise in health economic costs, we found that increases in GDP 75

per capita, population ageing, population growth, age-specific mortality reduction, and PM2.5 76 exposure changed the total health economic cost by 77%, 21.2%, 15.6%, -41.1% and -0.2%, 77

respectively. Compared to using an age-invariant VSLY or an age-invariant value of a 78 statistical life (VSL), the estimates of the older population's share of the total health economic 79 cost using an age-adjusted VSLY was 2 and 18 percentage points lower, respectively. 80 81

Although the negative health effects of air pollution increase with age, previous studies adopted 125 measures of the value of a statistical life (VSL) or value of a statistical life year (VSLY) that 126

were constant across all age groups to quantify the cost of air pollution attributable deaths. For 127

instance, the Lancet Commission on Investing in Health outlined a methodology to value the 128 changes in mortality by assuming that VSL was proportional to remaining life expectancy. 13 129

Estimates derived from this methodology do not account for the theoretical and empirical 130 evidence that both VSL and VSLY vary with people's age due to changes in remaining life 131

expectancy, life quality and socioeconomic status. 14-16 Therefore, applying an age-invariant 132 VSL or VSLY inaccurately estimates the health economic cost of air pollution, which might 133

bias the optimal allocation of resources for pollution control and public health investments. 134 135

In this study, we applied an age-adjusted VSLY to investigate the contribution of population 136

ageing to the global health economic cost of deaths attributable to ambient PM2.5 pollution. By 137

analyzing the interaction between ambient air pollution and population ageing, the findings 138 from this study will inform the design of pollution control policies and healthcare investment 139 strategies to improve the public health response to population ageing. 140 141 142

Research in context 143

Evidence before this study 144 We searched Web of Science, Google Scholar, and publicly available literature up to March 145 2020 for the terms "air pollution", "mortality", "health cost" and "population ageing" without 146

language restrictions to find studies that examine the relationship between population ageing 147 and the health economic cost of air pollution. Previous research found that population ageing 148 was a major driver of the substantial growth in global non-communicable diseases. However, 149 although many studies have assessed the health costs of air pollution, few have considered how 150 it changed as a result of population aging. We found multiple articles, including reports from 151 the World Bank, the World Health Organization (WHO) and the Organization for Economic 152

Co-operation and Development (OECD) that estimated the health cost of air pollution at both 153 regional and global scales. These studies applied either an age- We retrieved the global estimates of PM2.5 concentrations using a database that combines 202 satellite products and ground-based measurements with chemical transport modelling, 203 applying a geographically weighted regression (GWR) at 0.1  0.1 (~11 km at the equator) 204

resolution. 17 These estimates were more consistent with ground-based monitoring data than 205 PM2.5 estimates without a GWR adjustment.

To estimate the population that was exposed within each grid of PM2.5 concentration, we 208

collected population data (Gridded Population of the World -GPW v4) from NASA's 209

Socioeconomic Data and Applications Center (SEDAC), which is gridded at 0.0083  0.0083 210 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 5, 2020. . https://doi.org/10.1101/2020.04. 28.20083576 doi: medRxiv preprint (~1 km at the equator) resolution. 18 We aggregated the population data into the same (0.1  211 0.1) resolution as the PM2.5 data. 212 213

2. Estimating age-specific health risks from 2000-2016 214 215 We applied the GEMM to estimate the health risks from all non-accidental mortality due to 216 exposure from PM2.5 pollution. We also estimated deaths associated with five specific causes-217 chronic obstructive pulmonary disease (COPD), ischemic heart disease (IHD), stroke, lung 218 cancer (LC) and lower respiratory infection (LRI)-to understand their contribution to the total 219 mortality. 19 We defined non-accidental causes of mortality as all noncommunicable diseases 220

(NCD) and LRI. We subtracted the mortality related to the five specific causes of disease 221

(5COD) from all non-accidental causes of mortality (NCD + LRI) to compute the additional 222 non-accidental mortality (NCD + LRI − 5COD). We classified age groups in 5-year increments, 223

censoring the last age group at 85 years and older. The hazard ratio (here equated with relative 224 risk, or RR) of each cause of death k, at age m, in grid cell i, and year j is represented by the 225 following function: risk. We assumed that the counterfactual concentration follows a uniform distribution between 230 2.4 µg/m 3 and 5.9 µg/m 3 . 5 The functional form of GEMM is developed from the transformation 231 of a log-linear model. In the GEMM, and its SE control the slope of a non-linear regression 232 that represents the relationship between exposure concentration and RR, and defines the 233 curvature of the model. 5 We estimated attributable deaths using the following equation: 

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 5, 2020. , , refers to the exposed population, , , , is the years of life lost, and 256 , , is the life expectancy. We estimated attributable deaths and YLL with cause-specific, 257 location-specific, year-specific, and age-specific parameters. While attributable mortality and 258

population are observable at the grid cell level, we obtained the baseline mortality of each cause 259 and the life expectancy for each age group at the country level using data from the 2017 Global 260

Burden of Disease (GBD) Study. 27 This study explored the heterogeneity of health economic costs with respect to the age of the 265 exposed population by adopting an age-adjusted VSLY. Whether VSL and VSLY vary with 266

age has gained much attention in the literature. 28 To account for the variation in WTP over the lifecycle, we developed an age-adjusted VSLY 283 that modifies the constant VSL for wealth, remaining life expectancy, and age-specific survival 284 rates at the country level. First, we adjust for the effect of wealth over the lifecycle by 285 multiplying the constant VSL with age-specific wealth weights (wealth for certain age group 286 divided by the mean wealth of all-age group) that capture changes in consumption over a 287

lifetime. Individual wealth does not evolve monotonically with age. 36, 37 Resources and wealth 288 tend to accumulate before retirement age and then decrease modestly after retirement. 38 Figure  289 1a (see appendix pp 2) illustrates that the wealth ratio increases with age and decreases after 290 ages 65-70, while the ratio of aged population is higher than the wealth of the population 291 younger than 50 years old. We assumed that those under 18 years share the same wealth weight 292 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 5, 2020. . as their family members. This wealth-age profile suggests that older people, who have more 293

wealth to lose if they die, are likely to have a higher willingness to pay to avoid a unit of healthy 294

life year loss compared to younger adults with less to lose. 39 Second, we adjusted the constant 295 VSL by the ratio of the age-specific remaining life expectancy over the mean life expectancy 296

of the total population. The country-specific constant VSL was estimated using a benefit-297 transfer approach assuming a base VSL of US$3.54 million estimated from OECD 298 countries. 40, 41 This adjustment indicates that the VSL decreases with individuals' remaining 299 life years. Third, we adjusted the VSL by the survival probability to represent the quality of a 300

life year. Individuals with a low survival probability have little incentive to spend money to 301 avoid fatal mortality due to a low quality of life and few remaining life years. 39 

The value of age-adjusted VSLY is calculated from a remaining life-expectancy-and wealth-304

adjusted VSL and survival rate by age: 305

where _ is the age-adjusted value of a statistical life year, is the base VSL 307

from OECD countries, , is the gross domestic product (GDP) per capita for country c in 308

year j, , is the average GDP per capita of OECD countries, is the elasticity of 309 income in VSL, is the average wealth at age m, is the average wealth, T is the age 310

of expected death at age m, , is the survival probability at age m, and is the discount rate. The estimation procedure involved a range of parameters that have direct and indirect impacts 318 on the economic cost estimates. According to previous studies, the income elasticity of VSL is 319 about 1.0 for non-US countries. 43 Similarly, a meta-analysis of global VSL studies reports that 320 the income elasticity of VSL is between 0.9 and 1.3. 44 Following the recommendations of the 321

World Bank report, we selected a wide range of income elasticities for two categories of 322

countries. 8 For low-and middle-income countries, we select an income elasticity of 1.2, with 323 a range of 1.0 to 1.4 for our sensitivity test. For high-income countries, we select an elasticity 324 of 0.8 with a range of 0.6 and 1.0 for our sensitivity test. We set the discount rate at 6 percent 325

for low-and middle-income countries, and 4 percent for high-income countries. 8 The wealth 326 weights are calculated based on age-specific wealth distribution data extracted from national 327 statistical offices. Since the national wealth data of different age groups are very limited, we 328

only obtained age-wealth distribution data for five countries: the United States, Canada, the 329

United Kingdom, Germany and China. [45] [46] [47] [48] [49] [50] We then aggregated the country-level data into an 330 global average wealth data by age. We ran a loess regression model to simulate a global average 331 wealth weights in different age groups (appendix pp 1-2). 332 333 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 5, 2020. We estimated the economic cost within each age group as the product of the number of 337 prevalent YLL ( ℎ × ) in that group and the discounted present value of a life year loss. 338

In contrast to previous studies that used an age-invariant VSLY, we modelled the value of a 339

statistical life year at each age by its age-adjusted VSLY. where _ _ , , , represents the economic cost of PM2.5 using the age-adjusted 344 VSLY measure. 345 346

6. Uncertainty analysis and sensitivity test 347 348

Uncertainties in the distribution of PM2.5 concentration, exposed population sizes, life 349 expectancy, exposure-response functions, socioeconomic parameters and valuation methods 350

propagated to the health economic cost estimates. We adopted Monte Carlo simulations to 351 estimate 95% uncertainty intervals (UIs) from 1000 draws of parameters and exposure-352 response functions in the health economic cost assessment.

For our sensitivity analysis, we evaluated the health cost using four additional valuation 355 measures: i) a country-specific, constant VSLY across all age groups, ii) a global average age-356 adjusted VSLY, iii) a country-specific, age-adjusted value of statistical life (VSL), and iv) a 357 country-specific, constant VSL across all age groups. To compare the health economic cost 358 across countries and regions around the world, we also introduced a set of global average age-359 adjusted VSLY that removes differences due to variation in GDP per capita across countries. 360

All simulations were done with R version 3.6.0.

Role of the funding source 363

The funders of the study had no role in study design, data collection, data analysis, data 364

interpretation, or writing of the report. All authors had full access to all the data in the study 365 and ZL and QZ had final responsibility for the decision to submit for publication. 366 367 III Results 368 1. Disproportional health and economic cost on the elderly 369 370 Figure 1 illustrates the variation in age-specific mortality attributable to PM2.5 exposure. For 371 each additional unit of PM2.5 concentration, the average attributable mortality of people aged 372 60 and older was 12 times (95% UI: 11, 13) higher than the risks of those under 60. While a 373 decrease in ambient PM2.5 from 35 µg/m 3 (WHO interim target 1) to 15 µg/m 3 (WHO interim 374 target 3) leads to a 2.1% decrease in the attributable mortality rate among those age 85 and 375 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 5, 2020. . older, this rate decreased by only 0.6% in those 75-80 and it decreased even less in younger 376

ages. If PM2.5 was reduced to the WHO Air Quality Guideline (WHO AQG for PM2.5 is 10 377 µg/m 3 ), the attributable mortality for younger ages ranged between 0.02% and 0.1%. However, 378

the older population would still experience large risks, especially for those aged 70 and older 379 (0.4% to 1.6%). These estimates suggest that the impact of PM2.5 pollution will grow 380 substantially as the share of the population aged 60 and older expands. Therefore, the current 381 WHO AQG, which is determined by the average effects of air pollution on general population 382 without considering the effects of population ageing, might require strengthening. This 383 suggestion is further augmented by several recent analyses showing that significant health 384 impacts can still occur at levels of PM2.5 below the guideline. 51 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 5, 2020. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 5, 2020. . The increase in economic cost over this period was 20% faster in the older population than in the younger population. Globally, the health economic cost per capita of the older population was 10.2 times (8.1, 12.5) the per capita cost of the younger population. Hence, increases in air pollution disproportionately increases the economic cost with respect to the older population.

Over the 17-year period, PM2.5 attributable economic cost in the older population accounted for 57% to 59% of that in the total population. Importantly, despite generally lower air pollution concentrations, the High-income super region had the highest share of total economic cost borne by the elderly population (from 67% to 72%), followed by the Central Europe, Eastern Europe, and Central Asia super regions (from 56% to 60%). In comparison, given its much younger age distribution, only 24% of the total cost in the Sub-Saharan Africa super region was borne by the older population. The share of the health economic cost on the older population increased over time in most countries due to rapid population ageing ( Figures 2C  & 2D ). The proportion of economic cost related to the older population was higher in countries with relatively lower population-weighted PM2.5 concentrations as there was an overall negative correlation between PM2.5 levels and the national proportion of those older than 60 years (Appendix Figure 3a) . For example, Japan experienced the highest proportion of the health economic cost (from 70% to 83%) in the elderly from 2000 to 2016, followed by Italy (from 76% to 81%), Spain (from 71% to 77%) and Germany (from 70% to 76%) where the population weighted PM2.5 concentration was under 20 µg/m 3 .

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 5, 2020. 

The health damage attributable to PM2.5 varied substantially by age and cause. Among all the age groups, those aged 85 and older had the highest the number of deaths. Excluding the over-85 and under-20 age groups, mortality counts peaked at age 70-75 years with 0.85 million [0.67, 1.04] deaths, and YLL was largest for the aged 60-65 population with 14.89 million (11.48, 18.77) years lost ( Figure 3A . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 5, 2020. . https://doi.org/10.1101/2020.04. 28.20083576 doi: medRxiv preprint To account for differences in age structure between countries and to compare the health cost of PM2.5 across regions, we adopted a global average age-adjusted VSLY to estimate the cost of YLLs attributable to PM2.5 ( Figure 3D ). The results show that the largest health economic costs occurred in the South Asia super region and Southeast Asia and East Asia super region, which accounted for 58% (57%-60%) and 64% (63%-66%) of the global health economic costs in 2000 and 2016, respectively. Figure 3A illustrates attributable deaths by age across regions; Figure  3B represents the corresponding YLL in each age group and region; Figure 3C illustrates the variation in health economic cost by age and region over time using a country-specific and ageadjusted measure of VSLY; Figure 3D shows the distribution of health economic costs using a globally averaged, age-adjusted VSLY. Red error bars represent 95% uncertainty intervals.

Globally, attributable deaths and YLL associated with additional (beyond the 5 diseases included in the 2016 GBD study) nonaccidental causes (NCD + LRI − 5COD) accounted for 30% of attributable deaths and YLL from PM2.5. 54 Figure 4 indicates that additional nonaccidental mortality comprised 44% (36%-56%) of the total health cost attributable to PM2.5, . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 5, 2020. . though this varied widely by country. At the national level, the health economic cost of additional non-accidental mortality comprised over 30% of the total loss attributable to PM2.5 pollution in China and India in 2016. In comparison, the share of cost due to additional nonaccidental mortality in Japan and the United States was 56% (53%-64%). Overall, the North Africa, Asia and Central Europe regions had 30% of their total cost due to the additional nonaccidental mortality, while the High-income and the Latin America regions had more than 50% of their health cost caused by additional non-accidental deaths. From 2000 to 2016, highincome countries generally experienced a decrease in health economic costs due to the five specific causes, but an increase from the additional non-accidental mortality. On the other hand, low-middle-income countries had faster growth in economic costs associated with the five specific causes than that due to additional non-accidental mortality. The health economic costs of PM2.5 varied significantly among the five specific causes. IHD was associated with the highest costs among the five specific causes of disease, comprising 27% . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 5, 2020. . https://doi.org/10.1101/2020.04.28.20083576 doi: medRxiv preprint (26%-28%) of the total health economic cost attributable to PM2.5 in 2016. The cost of IHD was 5 to 10 times that due to the other four specific diseases. The global health costs associated with IHD increased from US$0.7 trillion (0.6-0.9) to US$1.1 trillion (0.9-1.4) from 2000 to 2016. The health cost associated with stroke increased fastest among all the specific causes, followed by COPD. The health economic cost of stroke in 2016 was 2.4 times that in 2000. More than 80% of the economic cost attributable to COPD was associated with the older population. This increased to an over 90% share in the aged countries, such as Japan, Italy, and Sweden. In contrast, countries such as the United Arab Emirates and Kuwait only had 30% of the economic costs caused by COPD related to the older population. In those older than 60 years, IHD, LC and stroke accounted for 63%, 68% and 64% of the total health economic cost attributable to PM2.5, respectively, in 2016. Figure 4 also reports the percentage of GDP loss due to the health economic costs related to PM2.5 pollution in 2016. China, USA, India, Russia and Japan were the countries that had the highest health economic costs worldwide. However, the health economic cost relative to national GDP was higher in Russia, India, Italy, Nigeria and China.

The substantial health cost of air pollution reflects the influence of environmental, demographic and socioeconomic factors. We decomposed the change in health cost attributable to ambient PM2.5 pollution from 2000 to 2016 by country and region into five major contributors-population growth, population ageing, age-specific mortality, the exposure level of ambient PM2.5 and the growth of GDP per capita (see appendix for detail method, pp 2-3). Globally, the effects of population ageing offset 52% of the benefits gained from mortality reduction over the 17-year period. Figure 5 shows the relative and absolute contribution of the five driving factors of the rise in the health cost of ambient PM2.5 in the total population ( Figure  5A ) and the older population ( Figure 5B ) in the GBD super regions. Among all driving factors, the increase in GDP per capita was the dominant contributor to the rapid growth of health cost over the study period in the South-East Asian regions ( Figure 5A-5B ). In the High-income super region, the change in age-specific mortality played a key role in reducing the health cost, lowering it by 28.5% between 2000 and 2016 ( Figure 5A ). However, the growth of health cost due to population ageing offset 90% of the benefits of avoided deaths due to the reduction of mortality in the High-income super region ( Figure 5A ). Except for the South Asia and Sub-Saharan Africa super regions, all the other regions experienced 18.8% to 33.1% growth in health cost due to population ageing ( Figure 5A ).

In terms of the older population, the contribution of population growth, population ageing, agespecific mortality and exposure of PM2.5 varied substantially ( Figure 5B ). The decrease in PM2.5 exposure benefited the older population more than the younger population. The effect of mortality reduction over this period avoided 56% of the health cost on the population under 60, while this reduction in mortality only brought a 29% decrease in health cost on the older population.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 5, 2020. . Figure 5A ) is the decomposition of the driving factors for the total population and the right one ( Figure 5B ) is for the older population.

The health economic cost of air pollution changes significantly with the valuation methods on which the estimates are based. Therefore, as sensitivity analyses, we compared the health cost estimates computed using the country-specific age-adjusted VSLY with those using: 1) a country-specific constant (age-invariant) VSLY, 2) a global average age-adjusted VSLY, 3) a country-specific age-adjusted VSL, and 4) a country-specific constant VSL, respectively. Unlike the country-specific methods, the global average age-adjusted VSLY does not place a higher value on the lives of people from richer countries when estimating the global health costs, we used a global average age-adjusted VSLY measure to compare the health cost at the global level.

Estimates of health cost were highest using the constant VSL measure compared to those using the other four valuation measures. In 2016, the health economic cost was US$8.32 trillion (6.62, . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 5, 2020. . 10.11) using an age-invariant VSL, which was more than two times the estimates based on an age-adjusted VSL (US$3.88 trillion [3.07, 4.73] ). In contrast, health costs were US$4.54 trillion (3.58, 5.57) and US$4.09 trillion (3.19, 5 .05) based on an age-invariant VSLY and an age-adjusted VSLY, respectively. The global average age-adjusted VSLY indicated that the global health economic cost was US$5.09 trillion (4.06, 6.19) in 2016, which was higher than the estimate from the country-specific age-adjusted VSLY.

Globally, the health economic costs were unequally distributed across various levels of exposure to PM2.5 concentration. This substantial variation was mainly due to differences in GDP per capita and annual PM2.5 concentration across countries worldwide. In 2016, the global health economic cost estimated by country-specific valuation measures (i.e., constant VSL, age-adjusted VSL, constant VSLY and age-adjusted VSLY) peaked at regions with PM2.5 concentrations of 14.5 µg/m 3 (Figures 6A, 6B, 6D, 6E ). Applying these four valuation measures, we also found that 50% of global health economic cost occurred in regions where PM2.5 was below 25 µg/m 3 . In contrast, if a global average age-adjusted VSLY measure is applied, half of the all-age health economic cost occurred in regions where ambient PM2.5 was below 41.5 µg/m 3 . For the constant VSL, the share of economic cost among the ageing population was 77%, while the share dropped to 60% when the age-adjusted VSL was used. In comparison, the older population's share of economic costs was 61% and 59% when we applied an age-invariant VSLY and an age-specific VSLY, respectively. Figure 5a (in appendix pp 6) shows that the United States, Japan and Russia, where a VSL was relatively higher due to the higher GDP per capita, had the highest health cost among the regions with PM2.5 levels below 25 µg/m 3 . In contrast, China and India had the highest health costs in the regions where annual PM2.5 was 25 µg/m 3 to 125 µg/m 3 , whereas these countries had a relatively lower VSL.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 5, 2020. . Figure 6 Global health economic cost distribution as a function of PM2.5 concentration by age group in 2016. Figures 6A-6E are area plots of the global age-specific health economic cost with the region level PM2.5 concentration. The integral of the area in each color represents the total health cost by age over a range of PM2.5 concentrations. Figure 6F illustrates the cumulative health economic costs with 95% UI, using four valuation measures and a synthesized valuation measure. Figure 6F presents the 95% confidence intervals of four country-specific valuation measures and a synthesized measure. For the synthesized measure, we aggregated the costs and uncertainties of these four measures by assuming that each valuation measure has an equal weight and value for policymaking. The health economic cost synthesized from the four country-specific valuation measures was US$5.2 trillion (4.1, 6.2), which was 27% higher than the estimates using the age-adjusted VSLY and 60% lower than the loss based on an ageinvariant VSL.

In this study, we examined the health economic cost of deaths by looking at the interaction between air pollution and global ageing. The population aged 60 years and older, which accounts for 10% to 12% of the global population, suffered 57%-59% of the total health economic cost of deaths attributable to PM2.5 over the 2000-2016 period, due to the combination of global ageing and the large impacts of air pollution on the older population. The health cost attributable to PM2.5 amongst this population was equivalent to 76% of the . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 5, 2020. . health spending in this age category in 2016, 55 which leads to aggravating challenge to national healthcare systems. The rate of increase in the economic cost in the older population was 20% faster than that in the younger population. The disproportional health costs with respect to the older people varied across regions and countries. The fraction of economic costs on the population aged 60 years and older was highest in the High-income super region and lowest in the Sub-Saharan Africa super region.

The rapid growth of global economic cost was mainly driven by the increase in GDP per capita, population ageing and population growth over the period of 2000-2016. The effects of population ageing offset 52% of the benefits attributable to the overall reduction in global mortality. In addition, the benefits of mortality reduction did not distribute evenly across all age groups. Reduction in mortality contributed to a faster rate of health economic cost abatement in the younger population, which might enlarge the disproportional economic cost on the older population. The growth in global health economic costs increased most rapidly in the Southeast Asia, East Asia and Oceania super region. This increase in health cost was mainly due to the growth of GDP per capita, increasing exposure of PM2.5 concentration and population ageing. In comparison, the increase of health cost in the High-income super region was mainly due to population ageing and growing GDP per capita.

The health economic costs attributable to ambient PM2.5 also varied significantly by cause. In addition to the five previously considered specific causes of disease in the 2016 GBD, the health impacts of additional non-accidental causes related to PM2.5 increased rapidly. Among the five traditional specific causes, IHD resulted in the highest per capita economic cost, five times that caused by LRI. Per capita health costs associated with lung cancer in the older population showed the highest average growth rate compared with that of other specific causes of disease. In comparison, the younger population had the highest growth rate in per capita economic cost associated with COPD. The differences of health economic costs by cause provide additional information to policymakers to facilitate the allocation of public medical resources. The additional costs of non-accidental causes of YLL suggest that the economic benefits of pollution control lead to much greater health benefits than previous estimates.

The health economic cost of ambient PM2.5 pollution fluctuates significantly according to the valuation measure used. The constant VSL measure generated the highest estimate of economic cost in 2016 (US$8.32 trillion [6.6, 10.1]) compared to the other four methods (ranging from US$3.07 trillion to US$5.57 trillion). Compared with an age-invariant VSL, which is used in most previous studies, the health economic cost of PM2.5 pollution using an age-adjusted VSLY measure considers three major effects that influence VSLY over one's lifetime: the change of remaining life expectancy, life quality and wealth weights. After the adjustment of VSLY by age, we obtained a set of VSL estimates that varied with age in an inverted-U shape that peaks at 40-50 years old (appendix pp 11), which is consistent with previous studies. Shepard and Zeckhauser (1984) illustrated in a "Robinson Crusoe" analysis that VSL increases with age among young adults, peaks at age 40, and then continuously drops when they get older. Simulation models of self-protection decisions that examine life-cycle consumption under agespecific mortality and dynamic wealth constraints also generated similar findings. 57 The life-. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 5, 2020. . cycle model of VSL depends on imperfect markets in which an individual is not allowed to borrow or lend money over his/her life time to avoid moral hazard problems. 58 Based on the hedonic model, estimated VSL peaks at age 46 years old, making an inverted-U shape, by observing market equilibria of tradeoffs between wages and fatal risks.

Although there are concerns about the inequality caused by adjusting VSL by age, social preferences revealed from consumers' choices and workers' wage rates imply that both VSL and VSLY peak at middle age and decline afterwards. 15 Therefore, an age-invariant VSL will lead to higher health economic cost estimates for the older population and an age-invariant VSLY will generate higher cost estimates for young people. 59 Since air pollution disproportionately affects the older population, individuals' risk-money tradeoff varies as their age, mortality rate and socioeconomic status change. 60 Applying the age-adjusted VSLY improves our knowledge on the health cost of mortality attributable to air pollution, especially in an ageing society. The health economic cost estimated using the age-adjusted VSLY facilitates the optimal allocation of pollution control resources and public healthcare investment for populations of different ages.

Despite the advancements to estimating the health cost of air pollution made in this study, limitations remain. First, we applied methods that require a series of global and historical data inputs, which are measured with uncertainty. Although the satellite-based estimates are generally somewhat lower than the ground monitoring data, the PM2.5 estimates used in this study show a highly consistency (R 2 =0.81) with ground measurements. 61 In addition, uncertainties in the population spatial distribution typically increase by the disaggregation of the age categories. Second, the exposure-response functions applied in this study also introduced uncertainties. Our estimates of economic costs caused by NCD+LRI were 40% higher than the total of five specific diseases in the older population using the GEMM. Due to the limited number of available cohort studies, the GEMM assumes that the prevalence of deaths for the additional causes, such as chronic kidney disease and dementia, 62, 63 is similar in all countries compared to the countries that are included in the 41 cohort studies. Including more cohort studies from highly-polluted countries and additional exposure-response functions for other NCDs in the GEMM will reduce the uncertainties in the estimates of deaths attributable to PM2.5 pollution. We also suggest that the exposure-response relationships of other NCDs should be further investigated to improve the estimates of health cost in additional NCDs. For example, the 2017 GBD Study included type 2 diabetes as another specific cause of disease, while the 2019 GBD study incorporated the estimates of deaths mediated by the impact of air pollution on birthweight and short gestation. Adding additional causes of disease related to PM2.5 might reduce the share of the health impacts of the older population relative to the total impact over all ages. In addition, as the GEMM is based on adult cohort mortality analyses, it may underestimate the overall impact of air pollution and specifically impacts on infant mortality and morbidity impacts on younger age groups. 64 Third, uncertainties embedded in the age-adjusted VSLY. We translated the VSL or VSLY of all countries worldwide based on the VSL database of OECD countries, because national studies are limited. However, this database might underestimate the VSL in the United States and even other higher-income countries. For example, Viscusi (2018) estimated that the VSL in the US is US$10 million based on a hedonic wage approach, which is much higher than the benefit-transfer estimate (US$ 4.5 million in the US) used in this study. If we adopt a VSL of US$ 10 million in the US, the health cost of PM2.5 in the US will be US$1.02 trillion (0.81, 1.22). Estimates of the income elasticity for VSL also varies across countries. Therefore, a benefit-transfer method induces uncertainties in VSL and VSLY estimates in countries. In this study, we applied an ageinvariant value of income elasticity within two groups of countries: low-income and highincome countries. To limit these uncertainties, we considered a wide range of elasticities between 0.6 and 1.4, which covers the estimates of elasticity reported in most previous studies. 40, 44 We acknowledge that elasticities are different across countries, but we believe our uncertainty assessment captured most of the bias caused by the variation of income elasticity in the VSL estimates. To adjust VSLY by age, we adopted a global average wealth weights by age due to limited statistical data at national level. More country-level wealth data by age will improve the robustness of the estimates in this study.

In conclusion, the health economic cost of deaths attributable to ambient PM2.5 among the older population accounted for a major share of total cost and has increased substantially from 2000 to 2016 owing to growing GDP per capita, population ageing and population growth. The health economic cost estimated by the age-adjusted VSLY measure provides further information for national and local government to design pollution control strategies and allocate healthcare resources by populations' ages. Additional non-communicable disease, not previously considered, represent a large share of the total health economic costs, thus the benefits of pollution reduction might exceed previous estimates. If substantial pollution reduction, especially in highly-polluted and highly-aged countries, is not achieved, ambient air pollution will lead to rapid increases in health economic cost related to mortality, which might induce substantial burden on national healthcare systems. In addition, improved health care targeted towards older individuals and providing opportunities to reduce exposures can be helpful to reduce the health cost of ambient air pollution. Given that many of the measures that will protect people from air pollution are the same ones necessary to decrease greenhouse gas emissions, this study provides a rationale for measures that will improve health across generations through both improved air quality and reduced health impacts related to climate change as the century progresses.

The detail of loess regression model is explained as following quadratic approximation:

To estimate the local regression value f̂(x 0 ), we solve the following equation to calculate the minimal β = ( 0 , 1 , 2 ) ′ :

where ( 0 ) = ( − 0 ℎ( ) ), ℎ( 0 ) is the span, here we applied a span of 0.3. Figure 1a presents the loess regression results. Although the wealth of people at old age decreased slightly, it remains to be higher than most of the younger age population. Correlation between population ageing and PM2.5 concentration We regressed the population ageing rates (proportion of population aged 60 years and older over the total population) with population-weighted PM2.5 concentration using a generalized additive model. Figure 3a shows that the ageing rates decreased continuously with an increase of population-weighed PM2.5 concentration, which represents an overall decreasing trend of population ageing with an incremental population weighted PM2.5 concentration. 8. Health economic cost distribution with PM2.5 concentration Figure 5a indicates the dramatic difference of health economic burden distribution with PM2.5 concentration by selected countries and the rest of the world. For high-income countries, such as the United States, Japan and Russia, the health cost mainly incurred in regions with PM2.5 concentrations below 25 µg/m 3 . By contrast, China and India suffered major economic loss with pollution concentrations between 25 µg/m 3 and 100 µg/m 3 .

Economic burden distribution with respect to annual average PM2.5 concentrations in the six specific countries and the rest of the world. 9. Premature deaths, YLL and health cost by age and income group from 2000 to 2016 Figure 6a -8a demonstrate the variation of premature deaths, YLLs and health economic cost by age groups from 2000 to 2016. The lower-middle income countries suffered the highest premature deaths and YLLs, whereas the health economic cost related to such health impacts were lower than the loss in high-income countries. This is mainly because in lower-middle income countries costs for health risk abatement are lower.

34 Figure 6a The attributable premature deaths from 2000 to 2016 by age in different income groups. 10. Health economic cost trends by cause and GBD super regions The health economic burden by causes and by regions are compared in Figure 9a . The Southeast Asia, East Asia, and Oceania region suffered 49% of global costs caused by COPD, followed by South Asia where it accounted for 21%. For lung cancer, Southeast Asia, East Asia, and Oceania and high-income regions experienced 47% and 36% of global health costs. High-income regions accounted for the highest proportion of total health costs caused by NCD-LRI and LRI. Southeast Asia, East Asia, and Oceania suffered the largest proportion of global health economic cost due to lung cancer, stroke and COPD. 11. Cumulative health economic cost by valuation measure Figure 10a illustrates the cumulative health economic burden distribution with respect to PM2.5 concentrations using five valuation measures. The age-invariant measure generates the highest estimates of economic cost related to PM2.5 pollution. With a global average age-adjusted VSLY measure, the estimates of economic burden grew substantially compared with country-& age-adjusted VSLY measure, which indicates that the most health damage happened in lower income countries. Figure 10a The growth cumulative health economic cost as a function of PM2.5 concentration by different valuation measures.

12. Cumulative health economic cost variations by age and measure To further compare the differences in estimates, we disaggregated the health economic burden distribution by different age groups using five valuation measures. It shows that the economic burden difference among the population younger than 60 years old was relatively consistent. However, the gap became larger at 60 and older, and especially large above 85. Specifically, the health economic burden valued by age-invariant VSL was 4.3 times the loss estimated by age-adjusted VSLY on the population aged 85 and higher. Figure 11a The comparison of cumulative economic burden distribution with PM2.5 concentration by age using five valuation measures. Figure 12a illustrates the effect of age on VSL over a life cycle period. In the age-adjusted VSL/VSLY measures, there were two inverted U-shape patterns after 20 years old, which shows that the VSL peaked at 40-50 years old (Figure 11a) , which is consistent with previous studies 4 . Since people younger than 20 years old generally have little cumulative wealth and there is no such data available for the analysis, we assumed the wealth weights are similar to their parents' wealth weights. For example, the wealth weight for people at 0-5 years old matched with wealth data of their parents aged around 25-30 years old. Therefore, the inverted U-shape pattern before 20 years old is due to the inverted relationship between age and VSL in the population at working age. Many empirical wagerisk studies revealed an inverted U-shaped pattern of age-VSL relationship 5, 6 . However, the hedonic wage models only revealed the WTP for mortality risk reduction during one's working age, i.e., in a rough range of 18 to 62 years old 4 . Due to limited studies reported the age-VSL relationship for people under 20 years old, we assumed that people under 20 years old have the same wealth as that of their parents. 

Revision 11

Estimates and 25-year trends of the global burden of disease attributable to ambient air pollution: an analysis of data from the Global Burden of Diseases Study

The contribution of outdoor air pollution sources to premature mortality on a global scale

Addressing Global Mortality from Ambient

Estimating health and economic benefits of reductions in air pollution from agriculture

A comparative risk assessment of burden of disease and injury attributable to 67 risk factors and risk factor clusters in 21 regions, 1990-2010: a systematic analysis for the Global Burden of Disease Study

Economic cost of the health impact of air pollution in Europe: Clean air, health and wealth

Interpretation of air pollution mortality: number of deaths or years of life lost?

New evidence on the impact of sustained exposure to air pollution on life expectancy from China's Huai River Policy

Institute for Health Metrics and Evaluation. Global Burden of Disease Study

A quantile estimation approach to identify income and age variation in the value of a statistical life

Age Differences in the Value of Statistical Life: Revealed Preference Evidence. Rev Environ Econ Policy

The impact of human health co-benefits on evaluations of global climate policy

Air pollution intervention and life-saving effect in China

Preferences for life saving programs: how the public discounts time and age

Diminishing Willingness to Pay per Quality-Adjusted Life Year: Valuing Acute Foodborne Illness

Admissible utility functions for health, longevity, and wealth: integrating monetary and life-year measures

Age effects in mortality risk valuation

The age-wealth relationship: a cross-section and cohort analysis

Population age structure, intergenerational transfer, and wealth: A new approach, with applications to the United States

The wealth-age relation among the aged

A handbook of Transport Economics -The value of a statistical life

The Income Elasticity of the Value per Statistical Life: Transferring Estimates between High and Low Income Populations

Valuing Mortality Risk Reductions from Environmental, Transport, and Health Policies: A Global Meta-Analysis of Stated Preference Studies

Income Elasticities and Global Values of a Statistical Life

Reliability of Meta-analytic Benefit Transfers of International Value of Statistical Life Estimates: Tests and Illustrations

Household wealth and finances in Germany

Household wealth and finances in Germany

China Household Finance Survey

Survey of Income and Program Participation

The impact of low concentrations of PM air pollution on mortality in the Canadian Community Health Survey-Mortality Cohort

Nonlinear associations between low levels of fine particulate matter and mortality across three cycles of the Canadian Census Health and Environment Cohort

Air pollution and mortality in the Medicare population

Global estimates of mortality associated with long-term exposure to outdoor fine particulate matter

Past, present, and future of global health financing: a review of development assistance, government, out-of-pocket, and other private spending on health for 195 countries

Survival versus Consumption

Explaining diversities in age-specific life expectancies and values of life saving: A numerical analysis

Policy challenges of the heterogeneity of the value of statistical life. Found Trends Microeconomics

Pricing Lives: International Guideposts for Safety

The heterogeneity of the value of statistical life: Introduction and overview

Global Estimates of Fine Particulate Matter using a Combined Geophysical-Statistical Method with Information from Satellites, Models, and Monitors

Estimates of the 2016 global burden of kidney disease attributable to ambient fine particulate matter air pollution

Air pollution and dementia: a systematic review

Robust relationship between air quality and infant mortality in Africa

Valuing mortality risk reductions from environmental, transport, and health policies: A global metaanalysis of stated preference studies

Global, regional, and national comparative risk assessment of 79 behavioural, environmental and occupational, and metabolic risks or clusters of risks in 188 countries, 1990-2013: a systematic analysis for the Global Burden of Disease Study

Estimates and 25-year trends of the global burden of disease attributable to ambient air pollution: an analysis of data from the Global Burden of Diseases Study

Age Differences in the Value of Statistical Life: Revealed Preference Evidence. Rev Environ Econ Policy

Adjusting the value of a statistical life for age and cohort effects

Estimating mortality risk reduction and economic benefits from controlling ozone air pollution

1. Country-specific age-invariant VSL and VSL estimation In the absence of empirical estimates of country-specific VSL, this study derived ageinvariant VSL by country using a benefit-transfer method based on meta-analysis for OECD countries 1 :where c represents country, j is the year.The discounted age-invariant VSLY is derived from dividing VSL by the remaining life years:where denotes individuals' life span, is the discount rate, represents the value of a statistical life, and is value of a statistical life year.

In this study, we generated five sets of economic cost estimates using four country-specific measures including age-adjusted VSL, age-adjusted VSLY, age-invariant VSL, ageinvariant VSLY and one global average age-adjusted VSLY measures.For loss of premature deaths, economic cost can be estimated using the loss of premature deaths multiplied by age-adjusted VSL or age-invariant VSL.where _ _ , , , represents economic cost of PM2.5 on the aged population using age-adjusted VSL measure.