Global and national assessment of the incidence of asthma in children and adolescents from major sources of ambient NO2

Pediatric asthma incidence has been associated with exposure to nitrogen dioxide (NO2) in ambient air. NO2 is predominantly emitted through fossil fuel use in land transportation, power generation and the burning of solid biofuels in households. We simulated NO2 with a global atmospheric chemistry model, combined with a land use regression model, to estimate NO2 exposure in all countries worldwide. The global asthma incidence among children and adolescents attributable to NO2 was estimated by deriving an exposure-response function from a meta-analysis which included epidemiological studies from multiple countries, baseline incidence rates from the Global Burden of Disease and gridded population data. The sectoral contribution to pediatric asthma from NO2 exposure (NO2-related asthma incidence: NINC) was estimated for different source categories to provide guidance to mitigation policies. We estimate 3.52 (2.1–6.0) million NINC per year globally, being about 14% of the total asthma incidence cases among children and adolescents. We find that emissions from land transportation are the leading contributor to NINC globally (∼44%), followed by the domestic burning of solid fuels (∼10.3%) and power generation from fossil fuels (∼8.7%). Biogenic emissions which are not anthropogenically induced may contribute ∼14% to the total NINC. Our results show large regional differences in source contributions, as the domestic burning of solid fuels is a main contributor to NINC in India and Nepal (∼25%), while emissions from shipping are the leading source in Scandinavian countries (∼40%), for example. While only 5% of all children and adolescents live in areas where NO2 exceeds the WHO annual guideline of 21.25 ppb (40 μg m−3) for NO2, about 90% of the NINC is found in regions that meet the WHO guideline, related to the uneven distribution of children and adolescents in the population. This suggests the need for stricter policies to reduce NO2 exposure, and revisiting the current WHO guideline to reduce the health risks of children and adolescents.


Introduction
Chronic respiratory diseases are among the leading contributors to morbidity worldwide (Soriano et al 2017, Stanaway et al 2018, Murray et al 2020. Asthma is a common respiratory condition leading to about 0.5 (95% confidence interval (CI) : 0.33-0.65) million premature deaths per year, associated with the loss of 22 (18-28) million disability adjusted life years in 2017 among all age groups (Soriano et al 2017, Stanaway et al 2018. Although the premature mortality from asthma decreased by 17.5% from 1990 to 2017, asthma affected about 330 million people in 2016) and according to the Global Burden of Disease (GBD) its prevalence increased by 30% in the same time period (Soriano et al 2017, Stanaway et al 2018. In low-and middle-income countries, where access to medical facilities and asthma medication is often limited, the increase is expected to continue in future (Foreman et al 2018, Global Asthma Network 2018.
Long term exposure to ambient air pollution not only exacerbates pre-existing asthma (Friedman et al 2001, Shima et al 2002, O'Connor et al 2008, Guarnieri and Balmes 2014, Brandt et al 2015, but is also a major driver of new asthma cases (Gauderman et al 2004, Jerrett et al 2008, Gehring et al 2013, Nishimura et al 2013, Guarnieri and Balmes 2014, Wijga et al 2014, Gehring et al 2020. Some studies indicate that air pollution increases the risk of the development of asthma during adulthood (Jacquemin et al 2009, Modig et al 2009, while infants (<5 years) are generally at greatest risk considering the immaturity of their lungs (USEPA 2008, 2016, Guarnieri and Balmes 2014. The human respiratory system develops fully in the first 18-20 years, and during this period there is a higher intrinsic risk for the development of asthma (USEPA 2016, Gehring et al 2020. The asthma incidence rate is about twice as high in young people (⩽19 years) than among adults (>19 years) (Soriano et al 2017, Stanaway et al 2018. The GBD has not yet included the impact of air pollutants on asthma. Recent studies demonstrate associations between both long-term and short-term ambient NO 2 exposure and asthma incidence in children, supported by experimental studies that characterize a probable adverse effect of NO 2 on the respiratory system (Gauderman et al 2004, Gehring et al 2013, Kudo et al 2013, USEPA 2016, Global Asthma Network 2018, Stevens et al 2019. The biological pathway of NO 2 exposure leading to airway inflammation, airflow obstruction and new asthma development is also well established (Fukuto et al 2012, Kim et al 2013, USEPA 2016, Stevens et al 2019. After assessing the evidence from recent cohort studies on exposure to NO 2 , the 2016 integrated science assessment report by the United States Environmental Protection Agency (US EPA) changed the formulation of causal determination from long-term exposure to NO 2 on respiratory effects from 'sufficient to infer a likely causal relationship' in the 2008 Integrated Science Assessment report to a 'causal relationship' (USEPA 2008(USEPA , p 201, 2016. The World Health Organization (WHO) sets the annual guideline for NO 2 at 40 µg m −3 (∼21.25 ppb at 25 • C temperature and 1 atmosphere pressure). However, this was proposed in the early 2000s when a robust basis for an annual average guideline for NO 2 through direct health effects was still lacking (WHO 2006).
Since the conversion between NO and NO 2 is rapid, typically they are considered as the tracer family NO x (Seinfeld and Pandis 2016). NO x is predominantly emitted as NO by anthropogenic activities involving the burning of fossil fuels (about 10% is directly emitted as NO 2 ), but it rapidly reacts with tropospheric ozone (O 3 ) to form NO 2 . Previous health impact studies used satellite retrievals and land use regression (LUR) models to estimate the exposure to NO 2 , and assumed the incident cases of asthma to be linked to transportation related emissions (Anenberg et al 2018, Achakulwisut et al 2019. While studies have shown that transportation is a major source of NO x , other emission sectors like biomass burning, agricultural waste burning, agricultural soils, power generation, industries, shipping and domestic solid fuel use also emit significant amounts of NO x (Gerstle et al 1965, Jaegle et al 2005, Tian et al 2013, Boersma et al 2015, Janssens-Maenhout et al 2015, USEPA 2016, Ding et al 2017, Degraeuwe et al 2019. Here, we combine the outputs from an atmospheric chemistry model and an LUR model to estimate the burden of incident cases of asthma among children and adolescents from the exposure to ambient NO 2 and attribute it to different source categories. Our study adds to previous work (Anenberg et al 2018, Achakulwisut et al 2019, Khreis et al 2019 by using an atmospheric chemistry model to distinguish the contributions by the abovementioned NO 2 sources, to guide policies aimed at regulating NO 2 exposure.

Model setup and emission inventory
We used the ECHAM5/MESSy atmospheric chemistry (EMAC)-general circulation model at a T106 horizontal spectral resolution (1.12 • × 1.12 • latitude × longitude), with 31 vertical hybrid terrainfollowing and pressure levels up to 10 hPa in the lower stratosphere (Sander et al 2005, Kerkweg et al 2006, Jöckel et al 2010, Joeckel et al 2016. The core atmospheric model is the 5th generation European Centre Hamburg (ECHAM5) general circulation model. EMAC includes sub-models that represent tropospheric and stratospheric processes and their interaction with oceans, land and human influences. We used the Modular Earth Submodel System (MESSy, v.2.54, (Joeckel et al 2010)) to link submodels that describe emissions, atmospheric chemistry, aerosol and deposition processes. The gas phase and heterogeneous chemistry was simulated through the Module Efficiently Calculating the Chemistry of the Atmosphere (MECCA) submodel (Sander et al 2011(Sander et al , 2019, which accounts for the photochemical oxidation of natural and anthropogenic emissions, also accounting for the volatile organic compounds. The evolution of organic aerosol compounds is simulated with the ORACLE submodel (Tsimpidi et al 2014(Tsimpidi et al , 2018, which effectively accounts for primary and secondary combustion products from biofuel, fossil fuel, biomass burning and their chemical oxidation during any atmospheric transport. Further, EMAC was nudged to simulate actual meteorological conditions, applying European Reanalysis Interim (ERA Interim) data for a period of 2 years (2014-2015), using 2014 as model spin-up.
We applied monthly varying sector-specific anthropogenic emissions to the model from the Community Emissions Data System (CEDS (Hoesly et al 2018)) at 0.5 • × 0.5 • resolution for the primary emitted species like sulfur dioxide (SO 2 ), oxides of nitrogen (NO x ), carbon monoxide (CO), black carbon (BC), organic carbon (OC), ammonia (NH 3 ) and speciated non-methane volatile organic compounds. We improved the model emissions over India using region-specific and updated emission inventories (Venkataraman et al 2020). Biomass burning emissions were obtained from the Global Fire Assimilation System (GFAS, Kaiser et al 2012). The emissions data were then pre-processed by distributing them in six emission heights; please see Pozzer et al (2009) for further details on distribution of emission from different sources to the six emission heights. Ambient NO 2 exposure was first estimated for all sources and then all source categories were removed one by one to obtain the contribution of nine sectors to ambient NO 2 exposure (Lelieveld et al , 2019a. The sectors considered are (a) land transportation (TRA) which includes emissions from road transportation and railways, (b) industries (IND) with emissions from iron and steel, paper and pulp, chemical, food, solvent and other manufacturing, oil refineries and fuel production, (c) domestic energy use (DOM) by the burning of fuels in households for cooking, lighting, domestic gas heating, diesel generators and biofuel use, (d) power generation (PGN) by thermal, oil and gas based power plants, (e) agricultural soils (AGS) that cause emissions from manure and fertilizer application and cultivation practices, (f) agricultural waste and residue burning (AWB), (g) emissions from ships and other water navigation (SHP), (h) biomass burning (BMB) which includes tropical, savanna, middle and high latitude forest fires, deforestation, savanna and shrub fires and (i) biogenic emissions from soils and plants (BGN).

Estimating the burden of incident cases of asthma
A recent study performed an extensive review of epidemiological studies that reported impacts of longterm exposure to NO 2 on NINC in children and adolescents (Anenberg et al 2018). Among five metaanalyses (Gasana et Khreis et al (2017) included 41 epidemiological studies performed mostly in East Asia, Europe and North America and reported an relative risk (RR) of 1.26 (95% confidence interval (CI) 1.10-1.37) per 10 ppb increase in ambient NO 2 . The information from (Khreis et al 2017) was utilized here to build an ERF for impacts of NO 2 on NINC using the following relationship: β was first estimated by taking ln-RR of 1.12 (95% Uncertainty Interval: 1.08, 1.16) for a 10 ppb increase in NO 2 . The β obtained was then used to obtain the RR for exposure to NO 2 using a low-concentration cut-off (LCC: 0.8 ppb). The LCC was selected based on the lowest concentration at which health impact was recorded (Khreis et al 2017(Khreis et al , 2019. We performed sensitivity studies with lower (0 ppb) and higher LCCs' (1 and 2 ppb). We evaluated the ERFs obtained from Khreis et al (2017) against those from the other four meta-analyses (Gasana et al 2012, Takenoue et al 2012, Anderson et al 2013, Bowatte et al 2015 and found that the estimates of NINCs from the ERFs drawn from all the five studies overlap within their uncertainty bounds (see: figures S1 and S12(b) (available online at stacks.iop.org/ERL/16/035020/mmedia)).
The burden of NINC among children and adolescents for 2015 was estimated using equation (2). A similar relation was applied in our earlier studies (Lelieveld et al 2019b, Chowdhury et al 2020 to estimate excess mortality where p is the exposed population with age ⩽19 years for 2015. The age profile of the population for 2015 was obtained from the GBD and was merged with population data at ∼1 km spatial resolution obtained from Global Human Settlement Layers (Florczyk et al 2019), resampled at 5 × 5 km spatial resolution. Age-specific baseline incidence rates (BINC) for the age classes <5, 5-9, 10-14 and 15-19 for asthma were obtained for each country from the GBD (https://vizhub.healthdata.org/gbd-compare/) . All estimates of NINC in this study are accompanied by 95% CIs. The 95% CI ranges were estimated by combining the uncertainty of the BINCs obtained from GBD and the uncertainty in the RR curves estimated by 10 000 random samples over the lognormal distribution of baseline incidence rates and RR. 95% CI was obtained from the resulting 10 000 estimates of NINC for each grid. Subsequently, the results were integrated and are presented at country level.

Modeled NO 2 exposure and bias-correction
The EMAC model simulated global, gridded NO 2 at 15 min time intervals has been averaged, and the output stored at monthly time periods. The spatial distribution of NO 2 EMAC is depicted in figure S2(a). The NO 2 EMAC was then evaluated against annual averaged in-situ observation data for 2015 (NO 2 OBS ), which were collected from ∼5000 sites around the world. The density of measurement locations is highest over Europe, East Asia and North America. NO 2 EMAC was found to have a low-bias compared to ground-based observations NO 2 OBS across the globe ( figure S2(b)). The global population weighted NO 2 EMAC was estimated at 4.3 ppb. The low-bias in NO 2 EMAC may be due to: (a) its simulation at coarse horizontal resolution (1.12 • × 1.12 • ) which averages out local gradients, this is especially relevant for urban areas and other locations with dense traffic, where mixing ratios are expected to be relatively high (below we will discuss the importance of traffic); (b), The lifetime of NO x (NO + NO 2 ) is short, typically of the order of a few hours to a day, hence spatial-temporal gradients between sources and locations downwind can be steep, which are not captured well at the applied horizontal and time resolution. However, the EMAC tropospheric column densities of NO 2 compared well with those from the ozone monitoring instrument (Boersma et al 2018), indicating that sources and sinks are overall well represented. Please see supplementary information sections S1, figure S3 for more details on the comparison.
The bias towards low values was most notable over densely populated regions, whereas in background locations EMAC performs rather well (figures S4(a) and (b), detailed description on identification of background sites across the globe is provided in supplementary information sections S2). To amend the low-bias in densely populated areas, we utilized a publicly available LUR model (Larkin et al 2017) with which a high-resolution NO 2 dataset was derived, henceforth referred to as NO 2 LUR . NO 2 LUR was established for 2011 at 100 m resolution globally, by using information from 5220 monitoring sites located in 58 countries globally and predictor variables like active fires, population density, proximity to major and minor roads, tree cover, impervious surface area, normalized difference vegetation index, water body and satellite retrieved NO 2 . NO 2 LUR was able to predict 54% of global variability, with a mean absolute error of 3.7 ppb (see Larkin et al (2017) for further details on formulation of NO 2 LUR ). The LUR model was shown to perform better compared to NO 2 EMAC over the densely populated and urban areas, and was able to capture the spatial variability over such locations. However, it was found to overestimate NO 2 compared to observations over the background locations (figure S5), with a mean bias (MB) of −44% for NO 2 OBS compared to NO 2 LUR . Also note the high values of NO 2 LUR over the regions with sparse availability of predictor variables and low monitoring station density (e.g. Saudi Arabia, Saharan Africa, Midwest USA, central Australia; see figure S5(a)). Sparse availability of predictor variables also impacted the efficiency of NO 2 LUR in biomass burning dominated regions of Central Africa (figure S5a).
To optimize the contributions of both NO 2 EMAC and NO 2 LUR , we combined the two data sets with information from the global human settlement layers settlement model (GHS-SMOD). The GHS-SMOD classifies each 1 km grid by population clusters into eight broad settlement classes: urban center, dense urban cluster, semi-dense urban cluster, suburban/peri-urban, rural cluster, low density rural, very low-density rural and water grid cells (see Florczyk et al (2019), Melchiorri et al (2018) for more details on GHSL-SMOD classifications). Merging of the data was performed by first resampling the NO 2 LUR at 1 × 1 km resolution and then implementing NO 2 EMAC over the background grids while using NO 2 LUR over the other grids. The grid cells with the GHSL-SMOD classifications low-density rural and very low-density rural were considered as background (more details on background grid classification in supplementary material section S2). A sensitivity study was performed by also including rural cluster grid cells into the background grid. Subsequently, the merged NO 2 exposure data were weighted by population (of children and adolescents ⩽19 years) to obtain NO 2 F (NO 2 EMAC fused ('F') with NO 2 LUR ) at 5 × 5 km resolution. Figures S6 and S7 depicts the algorithm for obtaining NO 2 F by combining NO 2 LUR and NO 2 EMAC by summarizing globally and also zooming in over a city.
The global population weighted NO 2 LUR was estimated at 9.8 ppb. Since 26% of the global child and adolescent population were estimated to reside in background grid locations (who were assigned NO 2 EMAC for exposure), the global population weighted NO 2 F was estimated to be lower at 8.1 ppb. The percentages of the population in all countries residing in background grid locations are depicted in figure S8(a). A large proportion of the child and adolescent population in Africa (31%) resides in background locations. In India, China, USA and Western Europe * 26%, 27%, 30% and 27% of the total child and adolescent population reside in background locations. We estimated the global population weighted NO 2 F exposure over background grid cells to be 2.9 ppb, compared to 6.3 ppb estimated with NO 2 LUR . The population weighted NO 2 F exposure at country-level is depicted in figure 1(a). We find that South Korea, Kuwait and Qatar have the * Countries considered in Western Europe: The Netherlands, Belgium, Germany, France, Austria, Portugal, Spain, Switzerland, Luxembourg, Italy, Republic of Ireland and the United Kingdom. Comparison of NO2 F with NO2 OBS at all observational sites. The area between the 1:2 and 2:1 lines is shaded gray. R is the correlation coefficient, MB the mean bias between NO2 OBS and NO2 F , and cp the percentage of measurement data that fall within the gray shaded region. (c) Global distribution of the pediatric population exposed to NO2. highest population weighted NO 2 F exposure at 18.6, 17.4 and 17.2 ppb, respectively. Population weighted NO 2 F exposure for China, the USA and India were estimated to be 12.5, 10.5 and 7.8 ppb, respectively. Figure 1(b) depicts the comparison between NO 2 OBS and NO 2 F across the world. The correlation coefficient for comparison between NO 2 OBS and NO 2 F (R = 0.66) is higher compared to comparison with NO 2 LUR (R = 0.63) and NO 2 EMAC (R = 0.43). A histogram of the pediatric population-NO 2 exposure curve was generated using the global gridded NO 2 F (figure 1(c)). The distribution was built by summing up the child and adolescent population in 500 logarithmically spaced bins from <1 ppb to 47 ppb and was found to have a distinct peak at 8 ppb. This feature is dominated by high exposure to NO 2 F among children and adolescents in India and China. The child and adolescent population-NO 2 exposure distribution of the five major countries are depicted in figure S9(a). We estimate that globally about 5% of all children and adolescents reside in areas where NO 2 F exceeds the WHO annual guideline level. These percentages are higher in South Korea (37%), Qatar (22%), Japan (16%), Egypt (12%) and China (10%) ( figure S9(b)), while only ∼2% of the children and adolescents in India and Western Europe live in areas where exposure exceeds the WHO guideline. We performed a sensitivity analysis by considering the GHSL-SMOD threshold to classify background grid cells by combining the rural cluster with the low density rural and very low-density rural grid cells as background locations and then merging NO 2 EMAC and NO 2 LUR to obtain NO 2 FR . About 36% of the global child and adolescent population lives in these (rural cluster, low density rural, very low-density rural) settlement grids ( figure S8(b)). Global population weighted NO 2 FR was estimated to be 7.8 ppb. Population weighted NO 2 F was higher than NO 2 FR for all countries, and figure S8(c) depicts the percentage difference between population weighted NO 2 F and NO 2 FR .

Burden of asthma incidence
Globally, we attribute 3.52 (2.16-6.05) million yr −1 NINC among children and adolescents to the longterm exposure to NO 2 , among which 56.5%, 18.1%, 16.4% and 10.8% occur among population of ages <5 years, 5-9 years, 10-14 years and 15-19 years, respectively. This amounts to ∼13% of the total incidence. The NINC is highest in China, followed by India, Brazil, the USA and Indonesia, in that order, (figure 2, table 1). The numbers for the top nine countries (ranked according to NINC) plus Western Europe are listed in table 1. The supplementary data present this information for all countries worldwide considered in this study. The total burden of NINC depends on the size of the exposed child and adolescent population, BINC and NO 2 F . For example, India has the highest number of exposed children and adolescents (520 million), being 1.75 times that in China; however, the lower BINC in India (425 per 100 000 children and adolescent) compared to China (917 per 100 000) and significantly lower population weighted NO 2 F (38% lower) in India result in 47%   S10(b)). Using NO 2 FR , we attribute 3.33 (2.06-5.78) million NINC to long-term NO 2 exposure. Using NO 2 LUR , 4.07 (2.5-7.01) million NINC (15.5% higher compared to NO 2 F ) was estimated, which is close to the result (4 million) obtained by a recent study that used the LUR model (Achakulwisut et al 2019). Figure S11 depicts the percentage differences between NINC estimated using NO 2 LUR and NO 2 F and NO 2 FR . Large differences (up to 55% in the sub-Saharan countries) were found between NINC calculated with NO LUR and NO 2 F in the countries with a large fraction of background grids (see figure S8). For China, India, Brazil and the USA the difference was estimated at 3.8%, 12%, 12% and 13.5% respectively.

Source sector attribution
The impacts of the nine aforementioned source categories was determined by omitting them one by one from the total NO x emissions in the global model results, following an earlier study . The relative difference between incident cases of asthma estimated by the simulation with total NO x and the simulation without emissions from a sector determines the contribution by the sector. The source contribution in the nine countries with highest NINC burden and Western Europe are depicted in figure 3 and listed in table 1. Land transportation (TRA) is the single largest contributor to NINC globally, accounting for 44%, followed by BGN, DOM, PGN, IND, AGS, SHP, BMB and AWB with 14.1%, 10.3%, 8.7%, 7%, 6.6%, 4%, 5% and 0.4%, respectively. The source contributions for all countries are presented in the supplementary data. TRA is the largest contributor to NINC in most countries in East Asia, Central Asia, Europe, North and South America (figure 4), associated with Figure 3. Contribution of the main sources of NO2 F to NINC in the nine countries plus Western Europe with the highest asthma incidence from NO2 F exposure.
1.56(0.95-3) million yr −1 NINCs. In Mexico and the USA, TRA contributes up to 73% and 59%, respectively. In Iran, Iraq, Saudi Arabia, Qatar and Kuwait, TRA is estimated to contribute to 60%-70% of the total NINC. We found it to be associated with 42%, 57% and 54% of NINC in China, Western Europe and Brazil, respectively. Its contribution is found to be lower in India (26%) and Nepal (29%) where the domestic use of solid fuels (DOM) is also a prevalent source of ambient NO 2 at the surface.
Power generation from fossil fuels (PGN) is a major source of NO 2 as well (Tian et al 2013). It is associated with 0.3 (0.18-0.51) million yr −1 NINCs globally. In Russia, the Philippines and China it is a major contributor to NINC with 25%, 23% and 21%, respectively. Its role is considerably less in the USA (6.3%), Western Europe (5%), Brazil (4.6%) and India (2.6%). Emissions from fuel combustion in industries and industrial processes (IND) contribute comparatively less. With a global fraction of ∼7%, it is associated with 0.24 (0.15-0.41) million yr −1 NINCs. It is relatively important in Japan (12%), South Korea (16%) and China (17.8%) while in most western high-income countries its contribution is comparatively low (4-8%).
The application of manure and fertilizers to soils for agriculture (AGS) is the next largest contributor at 6.6% accounting for 0.23 (0.14-0.4) NINCs. Use of nitrogen-based fertilizers has increased over time and is largely driven by intensive farming practices. In South Asia (e.g. India, Pakistan, Nepal, Bhutan and Bangladesh) where intensive farming is predominant, AGS contributes ∼20% to the total NINC. In Europe and North America, the contribution is small (up to 5%). Agricultural waste and residue burning is a minor contributor to NINC with 0.01 (0.006-0.02) million yr −1 . NO x from ship emissions (SHP) is expected to result in NINCs in coastal locations, major harbors and countries with coastlines adjacent to the major shipping routes. Globally, SHP is associated with 0.14 (0.09-0.23) million yr −1 NINCs. In Scandinavian countries it is a relatively large contributor to NINC, e.g. in Denmark (69%), Sweden (48%), Norway (44%) and Finland (20%). It also contributes significantly in island states, e.g. Iceland (65%), the UK (17%), Ireland (37%), Australia (16%) and New Zealand (14%).
Biomass burning, forest and grassland fires (BMB) account for 5% or 0.18 (0.11-0.31) million yr −1 NINCs globally. It plays a major role in sub-Saharan and particularly central African countries ( figure 4) where it can contribute more than 60%. It contributes ∼15% in Southeast Asia (e.g. Cambodia, Laos, Indonesia). BMB is also important (6%-10%) in South America and Australia. BGN from natural vegetation and soil is associated with 0.49 (0.31-0.87) million NINCs. It was found to be a major contributor to NINCs in sub-Saharan, central and south African countries (12%-20%). In China, USA, Brazil and India it contributes about 6%, 7.5%, 16.7% and 16.9% respectively.

Uncertainties and assumptions
All estimates of NINC in this study are accompanied by 95% CIs. The 95% CI ranges were estimated by combining the uncertainty of the BINCs and the uncertainty in the RR curves by distributing the RR and BINCs lognormally across the 95% CIs to obtain 10 000 estimates of NINC for each grid cell, from which the overall 95% CIs were estimated (see Chowdhury et al (2020) for details). Several other limitations may affect the results presented in this study. Firstly, the model was simulated to estimate NO 2 exposure at 1.12 • × 1.12 • horizontal resolution, which may be responsible for averaging out the local scale gradients in NO 2 exposure (USEPA 2016, Larkin et al 2017, Achakulwisut et al 2019. Nevertheless, this was alleviated by merging the model simulations (of 2015) with fine resolution LUR model calculations of NO 2 (of 2011). However, the source attribution was performed at the coarse model resolution and applied to the merged NO 2 F data, assuming only the steep NO 2 concentration gradients in the vicinity of sources, e.g. in urban centers, and not the emission strengths of the sources, which are not well represented.
Secondly, the main results of this study were obtained by using an LCC of 0.8 ppb, following the lowest concentration at which causality was noted in a recent analysis (Khreis et al 2017). We performed sensitivity calculations by considering no threshold (i.e. LCC = 0 ppb), and threshold levels of 1 ppb and 2 ppb (based on Achakulwisut et al (2019)) (figure S12(a)). The estimates of NINC were 12% higher and 3% and 16% lower when LCC was changed to 0, 1 and 2 ppb respectively (figure S12(a)). We estimated 3.93 (2.41-6.75) million, 3.42 (2.09-5.87) million and 2.95 (1.8-5.06) million incident cases of asthma from exposure to NO 2 per year among children and adolescents using 0, 1 and 2 ppb as LCC, respectively, which corresponds to 14.5%, 12% and 10.9% of the total global incidence of asthma, respectively.
Thirdly, the Khreis et al (2017) meta-analysis included epidemiological studies from East Asia, Europe and North America, therefore missing out on epidemiological evidence from the rest of the world. Also, an ERF derived from a single-pollutant epidemiological model was used to assume a causal relationship between NO 2 exposure and asthma incidence (because multipollutant models that adjust for other co-pollutants are currently not available). Nonetheless, the evidence is suggestive of more robust associations asthma incidence with NO 2 exposure, than for other air pollutants like PM 2·5 (Anderson et al 2013, USEPA 2016. However, it may be possible that the reported associations are sensitive to control for co-pollutants which may potentially change the estimates presented in this study. A sensitivity study was performed by estimating the NINC based on ERF curves from other available meta-analyses (figure S1). It was found that all the estimates of NINCs overlap within the uncertainty bounds (figure S12(b)).
A recent study by Holst et al (2020) that inspected association of multiple pollutants on onset of asthma found that only PM 2.5 remained significantly associated with asthma after controlling for other pollutants (NO 2 , PM 10 ) among children of parents with asthma and mothers who smoked during pregnancy. However, after adjusting for parental asthma, parental income and parental education, a significant association was obtained for NO 2 and new incidence of asthma among children. Another recent study (Gehring et al 2020) found stronger association of long-term exposure to NO 2 and incidence of asthma among children and adolescents compared to exposure to PM 2.5 . Nonetheless, we estimated the incident cases of of asthma from long-term exposure to PM 2.5 (PINC) simulated by the EMAC model as a sensitivity study to our estimates of NINC (supplementary information section S3). Globally, we estimated PINC to be ∼2 times higher than NINC, which is consistent with a previous study (Annenberg et al 2018). However we emphasize that, unlike for long-term exposure to NO 2 , there is no consensus among studies that associate exposure to long-term exposure to PM 2.5 with incidence of asthma (Gasana et al 2012, Anderson et al 2013, Guarnieri and Balmes 2014, USEPA 2016. We reiterate that our estimates of NINC and PINC are not additive considering significant overlap between them due to the use of single-pollutant epidemiological models in this study in absence of models that control for co-pollutants. We refer to supplementary information section S3 for further discussion.

Summary and conclusions
We estimate that 3.52 (1.2-7.2) million yr −1 new incident cases of asthma in children and adolescents are attributable to exposure to ambient NO 2 . Our result is slightly lower than that of a recent study (Achakulwisut et al 2019), which used exposure obtained from an LUR model only, but is three times that from a prior study (Anenberg et al 2018). We combined results from a global LUR model for NO 2 with atmospheric chemistry model simulations to obtain the NO 2 exposure metric. We estimate that globally 5% of all children and adolescents live in areas where NO 2 exposure exceeds the WHO annual guideline of 21.25 ppb (the WHO guideline of 40 µg m −3 was converted to ppb at standard temperature and pressure). However, ∼90% of the total NINC burden occurs in locations where the WHO annual guideline is met. This calls for revisiting the WHO annual guideline (WHO 2006) for NO 2 . It was formulated in 2006 when a robust basis for the guideline from epidemiological studies was not yet available (WHO 2006, USEPA 2008. If the areas where the WHO guideline for NO 2 is currently exceeded would actually meet the guideline, ∼2% [0.07 (0.04-0.11) million yr −1 ] of the NINC could be averted. Note that we do not account for the effect of NO 2 in causing exacerbations of symptoms in children and adolescents with preexisting asthma, and hence we provide a lower limit of the full effect.
To help direct policy decision making, we provide the first estimates of source sector specific contributions of NO 2 -induced NINC, whereas previous work attributed the exposure generally to traffic related sources (Anenberg et al 2018, Achakulwisut et al 2019. Nevertheless, our results corroborate the suggestion that land transportation (TRA) is the leading contributor to NINC, with a global average of 44%. In the USA and Mexico, the TRA contributions are very high, 58.8% and 73.4%, respectively. In Western Europe the contributions vary between 50% and 65%. In India and Nepal, where the burning of solid fuels in households is ubiquitous , Chowdhury et al 2019, this source sector is prevalent (∼25%). In China the industrial sector and power generation from fossil fuels together contribute ∼35% to NINC. In central Africa, biomass burning is the single largest contributor to NINC (>70%). Our results suggest that targeted air quality control measures and tightening of the WHO guideline for ambient NO 2 are needed to substantially reduce the burden of new asthma incident cases in children and adolescents.

Data availability statement
Data on incidence of asthma from exposure to NO 2 produced by this study are disseminated by country through SI Data1. Spatial raster of NO 2 F and data of incidence of asthma from exposure to NO 2 by provinces will be made available upon reasonable request to the authors.
The data that support the findings of this study are available upon reasonable request from the authors.