Ambient Air Pollution Exposure Association with Anaemia Prevalence and Haemoglobin Levels in Chinese Older Adults

Background: Health effects of air pollution on anaemia have been scarcely studied worldwide. We aimed to explore the associations of long-term exposure to ambient air pollutants with anaemia prevalence and haemoglobin levels in Chinese older adults. Methods: We used two-level linear regression models and modified Poisson regression with robust error variance to examine the associations of particulate matter (PM) and nitrogen dioxide (NO2) on haemoglobin concentrations and the prevalence of anaemia, respectively, among 10,611 older Chinese adults enrolled in World Health Organization (WHO) Study on global AGEing and adult health (SAGE) China. The average community exposure to ambient air pollutants (PM with an aerodynamic diameter of 10 μm or less (PM10), 2.5 μm or less (PM2.5), 1 μm or less (PM1) and nitrogen dioxide (NO2)) for each participant was estimated using a satellite-based spatial statistical model. Haemoglobin levels were measured for participants from dried blood spots. The models were controlled for confounders. Results: All the studied pollutants were significantly associated with increased anaemia prevalence in single pollutant model (e.g., the prevalence ratios associated with an increase in inter quartile range in three years moving average PM10 (1.05; 95% CI: 1.02–1.09), PM2.5 (1.11; 95% CI: 1.06–1.16), PM1 (1.13; 95% CI: 1.06–1.20) and NO2 (1.42; 95% CI: 1.34–1.49), respectively. These air pollutants were also associated with lower concentrations of haemoglobin: PM10 (−0.53; 95% CI: −0.67, −0.38); PM2.5 (−0.52; 95% CI: −0.71, −0.33); PM1 (−0.55; 95% CI: −0.69, −0.41); NO2 (−1.71; 95% CI: −1.85, −1.57) respectively. Conclusions: Air pollution exposure was significantly associated with increased prevalence of anaemia and decreased haemoglobin levels in a cohort of older Chinese adults.


Introduction
Anaemia is a significant public health problem in both industrialised and non-industrialised countries, with an estimated population prevalence of nearly one in four [1,2]. Anaemia is defined as that the increased cardiovascular morbidity and mortality associated with air pollution exposures may be partially explained by air pollution-induced anaemia [43].
Studies regarding the effects of air pollution on anaemia are scarce, with most of the prior research being focused on the short-term exposures or children population [26,43,44]. Only one study examined associations between air pollution and anaemia in older adults [45]. In that study, of 4121 older Americans, reported that an IQR increase in the one-year moving average PM 2.5 (IQR, 3.9 µg/m 3 ) and NO 2 (9.6 µg/m 3 ) were associated with a 33% (prevalence ratio, (PR) 1.33; 95% CI: 1.23-1.45) and 43% (PR 1.43; 95% CI: 1.25-1.63) increased prevalence of anaemia, respectively [43]. To our knowledge, no prior studies have examined whether air pollution is associated with anaemia or haemoglobin levels in an older Chinese population.
In our current study, we aim to address these gaps in the literature by examining the associations between long-term exposure to ambient particulate matter (PM) and gaseous pollutants (nitrogen dioxide (NO 2 )) with anaemia prevalence and haemoglobin levels in an older (>50 years) Chinese population. PM associations were examined for PM 10 (coarse particles with a diameter of less than or equal 10 µm), PM 2.5 (particulate matter with an aerodynamic diameter less than or equal to 2.5 µm) and PM 1 (particulate matter with an aerodynamic diameter less than or equal to 1 µm).

Study Population
The study on global AGEing and adult health (SAGE) is a longitudinal study run by the WHO, which aims to collect and analyse extensive nationally representative cohort data on the health and well-being of adult populations and the ageing process from middle-income countries, such as China, Ghana, Mexico, India, Russia and South Africa [46].
In our analysis, we used the data from the cross-sectional baseline survey acquired via interview from adult respondents from China during the 2007 to 2010 period (SAGE China Wave 1). A probability sampling design and a five-stage cluster sampling strategy were used. First, eight provinces from 34 eligible Chinese provinces were selected. Second, one rural county and one urban district from each of these eight provinces were randomly chosen. Third, two rural and two urban townships or communities were selected in each of the counties and districts from step 2. Fourth, from each of these townships and communities, two villages or enumerations were randomly chosen. Lastly, from each of the villages, two residential blocks were selected. All selections of the included geographic regions and districts in this multi-stage approach were performed randomly in order to minimise the possible risk of bias. Approval to conduct this study was granted through the Ethics Committee of the Chinese Centre for Disease Control and Prevention.

Exposure Assessment
Predicted data concentrations of PM 1 , PM 2.5 , PM 10 and NO 2 for China were modelled using satellite remote sensing, meteorology and land use information and other data combined with ground-monitored data on the pollutants from stations throughout China. Daily ground-level measurements of PM 1 were obtained from the 77 stations of the China Atmosphere Watch Network (CAWNET) from September 2013 to December 2014. Daily ground-level measurements of PM 2.5 , PM 10 and NO 2 were obtained from 1479 stations of the China National Environmental Monitoring Centre (CNEMC) from May 2014 to December 2016.
Details of the PM 1 predictions have been previously published [47]. Briefly, data on aerosol optical depth (AOD) was derived from two NASA Moderate Resolution Imaging Spectroradiometers (MODIS) data processing algorithms (i.e., Deep Blue and Dark Target), and then combined using an inverse variance weighting method at 0.1 deg (approximately 10 km) grid cell resolution. Ground-monitored PM 1 data was linked with the AOD data, vegetation data, land use information, meteorological data and other spatial data using a generalized additive model to predict daily PM 1 grid cell concentrations (2005 to 2014). A similar approach using random forest models was implemented to predict the daily PM 2.5 grid cell concentrations (2005 to 2016) and daily PM 10 grid cell concentrations (2005 to 2016) and the details have been published [48,49].
Likewise, details on NO 2 (from 2013 to 2016) prediction have also been published [50]. In summary, satellite-derived tropospheric column densities of NO 2 (molecules/cm 2 ) were obtained from the OMI-NO2 level-3 data product (OMNO2d version 3) [51] at 0.25 degree (approximately 13 × 24 km 2 ) resolution. Temporal convolution with Gaussian kernels, which tended to create gentle smoothing and preserve edges, was operated to filter out noise while filling data gaps because of satellite orbit, cloudy conditions, or surface reflectance. Ground-monitored NO 2 data was linked with satellite NO 2 data, vegetation data, land use information, meteorological data, road density and other spatial predictors using a random forest model to predict daily NO 2 grid cell concentrations.
Ten-fold cross-validation was performed using the monitored data to assess the predictive ability of the models for PM 1 , PM 2.5 , PM 10 and NO 2 . The results for the adjusted coefficient of determination (R 2 ) and root-mean-squared error (RMSE) are shown in Table S1.
We defined exposure to long-term ambient air pollution as the average PM and NO 2 concentrations of the three years (2005)(2006)(2007) prior to the survey. Finally, community locations of participants were geo-coded and then used to link the estimated annual PM and NO 2 concentrations.

Outcome Assessment
Using the WHO criteria, anaemia was defined as the haemoglobin concentrations below 13 g/dL in men and below 12 g/dL in women [3]. Haemoglobin was collected via dried blood spots captured and transported on filter paper to centralised laboratories. The analysis was performed using high-performance liquid chromatography. Haemoglobin from dried blood spots is tightly correlated with venous samples in population health studies [52,53].

Covariates
We controlled for demographic, biometric, health behaviour, indoor air pollution, socioeconomic status and comorbid disease covariates based upon their previous associations with either air pollution exposure [54,55], anaemia risk [56,57], or both [58]. The main potential confounders, such as age and sex, were included in the SAGE China Wave 1 survey. In addition to these, height and weight were also measured to calculate BMI. Two and more daily servings of fruits, as well as three or more daily serves of vegetables, were regarded as sufficient dietary intake [59]. Alcohol consumption and smoking status were defined as never or ever. The global physical activity questionnaire was used to measure the duration, frequency and intensity of general physical activity [60]. The questionnaire included information about moderate or vigorous physical activities during work, transport activities to and from places and recreation/leisure time activities. Based on the responses from participants, we classified them into three main groups based upon the intensity and time spent on each activity, corresponding to total energy requirements in metabolic equivalents (METs): low, moderate and high physical activity [60,61].
The types of fuel used at home for cooking were classified as clean (electricity and natural gas) and unclean (coal, wood, dung and agricultural residues). The information on the most common type of fuel for cooking was important for assessing indoor air pollution. The highest educational attainment was categorised as: (1) no school, (2) primary school and (3) middle or high school. The participants' socioeconomic status was categorised based on their self-reported household annual income: low ≤¥15,000; and high >¥15,000 using the median as the cut-off point. Finally, participants' self-reported hypertension, chronic lung disease and arthritis were used for comorbidity assessment.

Statistical Analysis
We examined the association between haemoglobin and an interquartile range (IQR) increases in three-year moving averages of PM 10 , PM 2.5 , PM 1 and NO 2 in single pollutant models. Haemoglobin measurements for participants in the same community may be correlated with each other, violating the independence assumption of regression models. To account this, we considered a two-level linear regression model where participants were considered as the first-level unit and the township as the second-level unit. For models, in which anaemia was treated as a dichotomous outcome, logistic regression may overestimate the prevalence ratio given anaemia's high prevalence in the study sample; therefore, we used a modified Poisson regression with robust error variance to estimate the prevalence ratio of anaemia [62]. Also, dose-response using mixed model cubic regression splines were calculated and plotted. Selection of covariates was based upon prior, established associations with air pollution and/or known biological effects on haemoglobin metabolism. In fully adjusted models, we adjusted for age, sex, BMI, tobacco use, physical activity, education level, fruit and vegetable intake, alcohol use, type of fuel used at home and median household income. We also investigated several potential effect modifiers age, sex, smoking, alcohol and region of the survey (north and south region). Because of spatial heterogeneity between the north and south areas of China in characteristics, such as, chemical components of PM, climate and some unknown confounders, which may modify the associations between ambient air pollution and anaemia. We conducted stratified analyses by dividing the survey region into north (including the other four provinces: Shandong, Hubei, Jilin and Shaanxi) and south (including the other 4 provinces: Yunnan, Zhejiang and Shanghai and Guangdong). Effect modification was assessed by including multiplicative terms between pollutant variables in single pollutant model and some potential effect modifiers into the models. Significance of effect modification was determined if the p-value for the hypothesis test of the interaction was <0.01.
We further estimated the anaemia burden attributable to ambient NO 2 by using three indicators: attributable cases, population attributable risk and population attributable fraction. The ambient NO 2 level (40 µg/m 3 ) set by the WHO Air Quality Guidelines was used as the reference concentration [63]. The population attributable fraction (PAF) were calculated using the formula below [64,65].
Here P b is the proportion of participants with anaemia, and OR is the prevalence ratio of anaemia in the study population. For ease of interpretation, we will present the PAF as population attribution risk percent (PAR%) by multiplying PAF by 100. To measure the adjusted contribution of NO 2 exposure above the WHO air quality guidelines on the total risk of having anaemia the 'adjusted PAR%' was calculated using the adjusted ORs from the final adjusted model. We estimated 95% confidence intervals for all PAR%.
A number of sensitivity analyses were performed to ensure our findings were robust to various model specifications. First, we used average pollutants concentrations for one and five years before the baseline survey. Second, we excluded the participants with comorbidities, since respiratory and cardiovascular diseases might have had an impact on the outcomes. All analyses were conducted using STATA version 15 (StataCorp, College Station, TX, USA) and p-value < 0.05 was used to determine statistical significance.

Results
Among 15,050 participants from the SAGE China Wave 1, 13,379 were 50 years of age or older. Of these, only 10,611 (79%) individuals had valid haemoglobin measurements; 645 additional participants were excluded because of missing covariate information, including sex, age, or type of fuel used at home. As a result, 9966 participants were included in our final analyses (Table S3). Anaemia was highly prevalent in the cohort, with 28.1% meeting the WHO definition. Participants with anaemia were significantly older than non-anaemic participants (64.3 versus 62.6 years, p < 0.01) were, exposed to higher NO 2 levels (32.3 versus 28.4 µg/m 3 , p < 0.01), had lower BMI (23.4 versus 24.3 kg/m 2 , p < 0.01) and were more likely to have lower education levels and use clean cooking fuels (p < 0.01). Non-anaemic participants were more likely to live in rural areas, have lower physical activity level and have sufficient intake of fruit and vegetables (Table 1). There was no significant difference in sex, smoking, alcohol intake and comorbidities between these two groups.  10 , particulate matter with an aerodynamic diameter less than or equal to 10 µm; PM 2.5 , particulate matter with an aerodynamic diameter less than or equal to 2.5 µm; PM 1 , particulate matter with an aerodynamic diameter less than or equal to 1 µm and NO 2 nitrogen dioxide. SD: standard deviation; BMI: body mass index.
The annual average of the four air pollutants varied across the study regions, with a range of 49-108 µg/m 3 for PM 10 , 29-69 µg/m 3 for PM 2.5 , 35-56 µg/m 3 for PM 1 and 20-46 µg/m 3 for NO 2 . The annual average of PM 10 and PM 2.5 concentrations in the study regions were higher than the WHO guidelines in all eight regions of the SAGE China study (Table 2).  Table 3 shows the associations between PM and NO 2 and anaemia prevalence ratios. For PM 10 and PM 2.5 , univariate models showed non-significant associations. After adjusting for potential confounding factors, these associations became positive and significant, with an IQR increase in PM 10 Table 4 shows the statistically and clinically significant associations between PM and NO 2 and haemoglobin levels. For each PM 10  (respectively), after which they become positive. In contrast, for PM 1 the strongest negative associations occur at the lowest and highest exposures, with a curvilinear association for intermediate exposures.
For NO 2 , no association is observed until 40 ug/m 3 , which is the WHO air quality guideline for NO 2 , after which they become strongly and significantly negative.  In stratified models, the association between ambient PM and NO 2 and haemoglobin (Table S4) suggest that smoking may modify the association between PM exposure and haemoglobin levels, but not NO 2 . In PM 1 and PM 10 models, we observed a higher decrement of haemoglobin per IQR increase in smokers (−0.78 ± 0.10 vs. −0.61 ± 0.08 (p = 0.003) and −0.80 ± 0.10 vs. −0.58 ± 0.07 (p = 0.006), respectively), suggesting that smoking may enhance this association. Also, alcohol consumption showed a significant effect modification of this association for PM 1 . We observed a larger magnitude of decrease in participants who reported alcohol consumption (−0.69, β 95% CI: −0.80, −0.58, p < 0.001) than for those who reported that they never consumed alcohol (−0.61, β 95% CI: −0.68, −0.54, p < 0.001). Region of survey was also found to be an effect modifier, participants living in northern province showed higher decrement of haemoglobin per IQR increase levels compared to southern residence (−0.29 ± 0.06 vs. −0.04 ± 0.20 p < 0.001) respectively. We did not observe any significant effect modification by age, sex, household income or history of chronic obstructive pulmonary disease (COPD).
Estimates of the anaemia burden attributable to ambient NO 2 after adjusting for confounders are summarised in Table 5. The population attributable risk associated with ambient NO 2 higher than 40 µg/m 3 was 4.4% (95% CI: 3.6-5.2) corresponding to 131 (95% CI: 107-155) anaemic cases in the study population. Further, we estimated the PAF and observed that approximately 26% of the anaemic cases in the study could have been averted if all older adult participants lived in areas where NO 2 levels met the WHO air quality guidelines, assuming a causal association (Table 5). The associations between PM and NO 2 and anaemia did not materially change in the sensitivity analyses (Table S5). For example, when we used one-year and five-year average PM 1 concentrations as the exposure variable, the PR for each IQR µg/m 3 increase were 1.12 (95% CI: 1.05-1.21) and 1.15 (95% CI: 1.07-1.24), respectively. Additionally, when we excluded the respiratory and cardiovascular cases from the data, the associations remained robust, with PRs of 1.26 (95% CI: 1.17-1.36) per IQR increase in PM 1 .

Discussion
The current study is the first to show that long-term exposures to PM and NO 2 are significantly and consistently associated with decreased haemoglobin levels and increased prevalence of anaemia in a large scale survey of older adults in China. In addition, we observe that among current smokers, the adverse effects of PM 10 and PM 1 , but not NO 2 , on haemoglobin levels were even stronger (p < 0.01). Moreover, alcohol consumption was shown to enhance this link for PM 1 and NO 2 (p < 0.01), but not for PM 2.5 or PM 10 . Dose-response curves for all associations demonstrated significant non-linearity. We further estimated that 4.4% of anaemia cases in the study population could be attributed to NO 2 exposure and that 26% of the anaemia cases observed in the study could have been averted if NO 2 in the study locations met the WHO guidelines.
Our findings of the air pollution impact on haemoglobin levels in older adults are consistent with the results of a recent US study. Honda et al. found that an IQR (3.9 µg/m 3 ) increase in one-year moving average PM 2.5 was linked to significantly higher anaemia prevalence (PR 1.33; 95% CI: 1.23-1.45) and decrement in haemoglobin concentrations of 0.81 g/dL (95% CI: 0.75-0.87). Similarly, an IQR (9.6 ppb) rise in NO 2 was associated with an increased prevalence of anaemia (PR 1.43; 95% CI: 1.25-1.63) and a decrease in average haemoglobin of 0.81 g/dL [43]. Our findings are also consistent with other studies evaluating the associations between long-term air pollutant concentrations with haemoglobin levels in a younger population. A 2006 study of 327 Serbian individuals found that a higher five-year average pollution (sulphur dioxide (SO 2 ), soot and lead (Pb) in sediment matters in the air) was linked to an increased risk of anaemia during pregnancy [66]. Meanwhile, Nikolic' et al. found that children who are exposed to a high level of long-term NO 2 , black smoke and SO 2 had lower haemoglobin levels and RBC counts compared to those who lived in less polluted areas [67].
Studies investigating associations between short-term air pollution exposures and haemoglobin levels show inconsistent results. For instance, in a UK study of acute air pollution exposures in 112 older adults, three-day average PM 10 exposure was associated with an 0.44 g/dL decrease in average haemoglobin per 100 µg/m 3 increase in PM 10 [58]. Similar findings were reported in a cross-sectional study in India, in which 24-hour exposures of PM 10 , NO 2 and SO 2 were significantly associated with lower haemoglobin concentrations (p < 0.05) [68]. In contrast, a study by Sørensen et al. of 50 young adults (aged 20-33 years) found that PM 2.5 and black carbon exposures were associated with higher haemoglobin and erythrocyte counts, with a two-day average air pollution exposure associated with a 2.3% increase in RBC number, and a 2.6% increase in haemoglobin per 10 µg/m 3 increase in PM 2.5 in women, but not in men. In a study of children aged 6-59 months living in one of the most air-polluted cities of Latin America (Lima, Peru), investigators found that high levels of outdoor PM 2.5 (25-50 µg/m 3 ) were associated with a 39.6% of anaemia prevalence (95% CI: 39.3-39.9), were conversely males were one of the most affected groups [69].
The inconsistency of the findings from the studies mentioned above on the short-term air pollution exposures may reflect several factors. First, all these studies have a small geographically distinct and thus non-representative sample size, which limits the generalisability of the findings to other populations. Second, short-term pollutant exposure likely affects haematologic and haematopoietic physiology differently than long-term exposures [70], and older adults are possibly affected differently than younger individuals.
Air pollution is known to cause a chronic systemic inflammatory response [71]. This, in turn, may lead to decreased erythropoietin secretion by the kidneys as well as increased endogenous erythropoietin resistance in the bone marrow [72,73]. Together, these two factors can lower the production of RBCs and result in decreased haemoglobin levels [72]. Another possible explanation is that air pollution can induce IL-6 gene expression in alveolar macrophages, which has been previously shown to upregulate hepatic hepcidin production [74]. Increased hepcidin, an iron regulatory protein, can decrease gastrointestinal iron absorption, contributing to iron deficiency anaemia [75]. These mechanisms related to chronic inflammation and perturbed iron regulation, taken together, might explain how long-term, although not short-term, air pollution exposures may contribute to decreased haemoglobin levels and anaemia [43].
We found that for PM 10 and PM 2.5 , the adverse effects of air pollution on haemoglobin levels were amplified in smokers (−0.78 ± 0.10 g/dL) as compared to non-smokers (−0.61 ± 0.08 g/dL). This finding is inconsistent with findings by Honda et al. (2018), who found that NO 2 exposure, but not PM, was associated with higher magnitude associations in older US adults. It is possible that local immune response differences between smokers and non-smokers may underlie our result, which makes smokers more susceptible to air pollution exposures. Additionally, effect modification was observed for alcohol consumption, where increases in PM 1 and NO 2 among those who reported ever consuming alcohol were associated with higher magnitude associations relative to those who reported never consuming alcohol. This outcome is also unexpected, as prior literature has found that alcohol consumption is often associated with increased haemoglobin concentrations [76]. Also, we found greater association in the northern regions of china compared to south of china, this outcome is expected because of the spatial distribution of ambient air pollution in China. Previous studies reported the average concentration of PM 2.5 in north China reached as high as 80-100 µg/m 3 , while in south, the concentration has reduced to 40-70 µg/m 3 [77][78][79].
Our findings that PM and NO 2 were associated with decreased haemoglobin levels in Chinese older adults have significant clinical and public health implications. For instance, a 2005 study of older adults (mean 60.2 years) post-myocardial infarction found that just 1 g/dL decrease in haemoglobin levels below 14 mg/dL was linked to 21% increased odds of cardiovascular mortality (95% CI: 12-30%) [80]. This study evaluated the anaemia burden attributable to ambient NO 2 in a population of Chinese elderly. It also estimated the reduction of the anaemia burden when the NO 2 levels meet the WHO air quality guidelines. Most of the previously published works usually quantified the association between pollutants and adverse health outcomes. The attributable risk may provide even more information from the point of public health significance. A similar approach has been applied to other research works. In a recent study of six low-to-middle income countries, Guo et al. estimated that 7.11% of stroke cases could be attributable to ambient PM 2.5 [81]. Moreover, the Global Burden of Disease study reported that 4.2 million (7.6%) deaths globally could be prevented if the standards of the ambient PM 2.5 were attained [82]. The latter suggests that the impact of air pollution on anaemia may be one way in which air pollution leads to cardiovascular mortality [83].
Our study has a number of important limitations. First, as a cross-sectional study, we are not able to establish a temporal relationship between the exposures and outcomes of interest. Second, while we used a satellite-based comprehensive exposure model and assigned exposures according to the participants' addresses at the township level, the possibility of exposure misclassification is high, as we were unable to have more detailed information to inform our exposure assessment, such as the respondents' activity patterns regarding time spent in traffic and indoors. Third, specific components of PM that may be driving our results, such as transition and/or heavy metals, could not be considered separately. They might have had distinct chemical structures and might be associated with different health concerns. Last, some critical risk factors of anaemia, such as genetic background and cultural dietary differences among various geographic areas, were not taken into account. However, in our multilevel models, we attempted to address this by controlling for the community in which an individual lives. Several significant strengths counterbalance these limitations. First, our study encompasses a large, representative population to examine the association between ambient air pollution and anaemia among older adults in China. We collected extensive individual-level information on other risk factors, allowing us to control for individual-level confounders. Second, the use of satellite-based estimates of PM 1 and PM 2.5 exposures enabled us to have a virtually complete spatial coverage among the study participants in the absence of air monitoring data in China for PM 2.5 before the year of 2012.
Third, our study is the first to investigate the PM 1 association with anaemia in any settings globally. The existing air pollution standards in both the United States and China do not include regulations for PM 1 . Therefore, there is an urgent need to fill these gaps in the literature and our study can be used to inform public policies and guidelines in order to protect older adults from hazardous risks associated with PM 1 air pollution.

Conclusions
Our study is the first to demonstrate that exposure to ambient air pollution is associated with an increased prevalence of anaemia and reduced levels of haemoglobin in Chinese older adults. Further studies to examine the mechanisms accounting for increased vulnerability to the PM and NO 2 are warranted. Public policies and guidelines should be improved to protect the ageing population from risks associated with air pollution.
Supplementary Materials: The following are available online at http://www.mdpi.com/1660-4601/17/9/3209/s1, Table S1: Results of 10-fold cross-validation for PM 1 , PM 2.5 , PM 10 and NO 2 . Table S2: Annual average concentrations and pair wise correlations of air pollutants. Table S3: Characteristics of participants with haemoglobin blood samples data and without blood sample data. Table S4: Associations between 3-year pollutant moving averages and haemoglobin, by participant characteristics. Table S5: Sensitivity analysis of the association between anaemia prevalence and moving averages of pollutants. Figure S1: the association between an 10 ug/m 3 increase of PM and NO 2 and decrements in haemoglobin (g/dL). Funding: "This research was funded by CAR (Centre for Air quality and health Research and evaluation)." and "School of public health University of Sydney".