DSpace at VNU: Effect of poverty on the relationship between personal exposures and ambient concentrations of air pollutants in Ho Chi Minh City

The objective of the study was to characterize the levels, determinants of exposure, and relationships between children personal exposures and ambient concentrations of multiple air poll

Effect of poverty on the relationship between personal exposures and

ambient concentrations of air pollutants in Ho Chi Minh City

Sumi Mehtaa, Hind Sbihib,*, Tuan Nguyen Dinhc, Dan Vu Xuand, Loan Le Thi Thanhe,

a Health Effects Institute, Boston, MA, USA

b School of Population and Public Health, University of British Columbia, 2206 East Mall, Vancouver, BC V6T 1Z2, Canada

c Ho Chi Minh City Environmental Protection Agency (HEPA), Institute for Environment and Resources (IER), The National University of Ho Chi Minh City,

Viet Nam

d Center for Occupational and Environmental Health, Viet Nam

e Ho Chi Minh City Bureau of Statistics, Viet Nam

f Ho Chi Minh City University of Science, Viet Nam

g Department of Public Health, Viet Nam

h i g h l i g h t s

We examined the pollutant exposureepoverty relationship in Ho Chi Minh, Vietnam

Personal exposures to particles and NO2were higher amongst the poor

Ambient levels poorly reflect personal exposures, in particular for poor residents

In addition to socioeconomic status, behavioral factors determined exposure levels

a r t i c l e i n f o

Article history:

Received 15 April 2014

Received in revised form

30 June 2014

Accepted 3 July 2014

Available online 3 July 2014

Keywords:

PM

NO 2

Asia

Socioeconomic status

Exposure assessment

a b s t r a c t

Socioeconomic factors often affect the distribution of exposure to air pollution The relationships be-tween health, air pollution, and poverty potentially have important public health and policy implications, especially in areas of Asia where air pollution levels are high and income disparity is large The objective

of the study was to characterize the levels, determinants of exposure, and relationships between children personal exposures and ambient concentrations of multiple air pollutants amongst different socioeco-nomic segments of the population of Ho Chi Minh City, Vietnam Using repeated (N¼ 9) measures personal exposure monitoring and determinants of exposure modeling, we compared daily average

PM2.5, PM10, PM2.5absorbance and NO2concentrations measured at ambient monitoring sites to mea-sures of personal expomea-sures for (N¼ 64) caregivers of young children from high and low socioeconomic groups in two districts (urban and peri-urban), across two seasons Personal exposures for both PM sizes were significantly higher among the poor compared to non-poor participants in each district Absolute levels of personal exposures were under-represented by ambient monitors with median individual longitudinal correlations between personal exposures and ambient concentrations of 0.4 for NO2, 0.6 for

PM2.5and PM10and 0.7 for absorbance Exposures of the non-poor were more highly correlated with ambient concentrations for both PM size fractions and absorbance while those for NO2 were not significantly affected by socioeconomic position Determinants of exposure modeling indicated the importance of ventilation quality, time spent in the kitchen, air conditioner use and season as important determinant of exposure that are not fully captured by the differences in socioeconomic position Our results underscore the need to evaluate how socioeconomic position affects exposure to air pollution Here, differential exposure to major sources of pollution, further influenced by characteristics of Ho Chi Minh City's rapidly urbanizing landscape, resulted in systematically higher PM exposures among the poor

© 2014 The Authors Published by Elsevier Ltd This is an open access article under the CC BY-NC-ND

license (http://creativecommons.org/licenses/by-nc-nd/3.0/)

* Corresponding author.

E-mail addresses: hind.sbihi@ubc.ca , hind.sbihi@gmail.com (H Sbihi).

Contents lists available atScienceDirect Atmospheric Environment

j o u r n a l h o m e p a g e : w w w e l s e v i e r c o m / l o c a t e / a t m o s e n v

http://dx.doi.org/10.1016/j.atmosenv.2014.07.011

1352-2310/© 2014 The Authors Published by Elsevier Ltd This is an open access article under the CC BY-NC-ND license ( http://creativecommons.org/licenses/by-nc-nd/3.0/ ).

Atmospheric Environment 95 (2014) 571e580

1 Introduction

Asia is undergoing significant economic development,

popula-tion growth, and urbanizapopula-tion with subsequent industrializapopula-tion

and growth in vehiclefleets leading to increased emissions of air

pollutants and shifts in environmental risks (HEI International

Scientific, 2010) As a result, large populations in rapidly

devel-oping economies of Asia are exposed to high concentrations of air

pollution These exposures, coupled with aging populations and

increasing burden of chronic diseases, have led to substantial

population health impacts from air pollution The recent Global

Burden of Disease estimated over 2.1 million premature deaths and

52 million years of healthy life lost in Asia from ambientfine

par-ticle air pollution in 2010, 2/3 of the worldwide burden In

South-east Asia, the region which includes Vietnam, outdoor air pollution

was estimated to contribute to 712,000 deaths in 2010 (Lim et al.,

2012; Wang et al., 2012)

The public health and social policy implications of the

re-lationships between health, air pollution, and socioeconomic

po-sition are likely to be important in Asia, where air pollution levels

are high and many still live in poverty Despite what appears to be a

similar magnitude of population risk for a given level of exposure to

air pollution (Wong et al., 2010, 2008) there is still a lack of

evi-dence about exposure sources and determinants in urban Asia

compared to North American and Europe Economic deprivation

has been shown to increase the rates of morbidity and mortality

related to air pollution in Europe and North America (Finkelstein

et al., 2005; Laurent et al., 2007), and socioeconomic status

dic-tates the vulnerability of population to environmental risks via

factors such as nutritional status and access to medical services In

Asia where large income disparities are more prevalent than in

many high-income countries, results of Western studies cannot

merely be extrapolated Variation in socioeconomic status within

Asian populations could impact exposures differently than in

developed countries, particularly for determinants related to urban

planning (residential location, proximity to traffic and small-scale

industries), as well as lifestyle and time activity patterns

Expo-sure to indoor combustion sources in the Asian context (for

example from incense use and cooking) also differ from those in the

Scientific., 2010; Le et al., 2012; Smith et al., 2000)

Studies of personal exposure conducted in developed countries

indicate that for time series studies of the effect of daily change in

air pollution levels, central monitoring sites are adequate

surro-gates for longitudinal changes in exposures (Janssen et al., 1998,

2005; Sarnat et al., 2000) To date no studies of this type have

been conducted in many of the poorer Southeast Asian countries,

such as Laos, Cambodia, and Vietnam While it is possible to apply

existing studies from developed countries to help tailor air quality

management strategies, there is a need to assess the extent to

which localized sources, time activity patterns, and socioeconomic

position may contribute to exposure estimation in the Asian

context

Under an initiative of the Asian Development Bank, an

inter-disciplinary collaboration between local and international experts

launched assessed the health effect of air pollution and the role of

poverty in Ho Chi Minh City (HCMC), Vietnam An epidemiologic

study was conducted to evaluate the impact of air pollution on

childhood respiratory infections (children<5 years) between 2003

and 2005 (Le et al., 2012; Mehta et al., 2013) Thisfirst study of the

health effects of air pollution in HCMC suggested a potential role of

air pollution exposure (in particular NO2) in the development of

Acute Lower Respiratory Infections (ALRI), but was unable to

identify differential effects by socioeconomic position likely due to

the small number of patients identified as poor Given uncertainties about the extent to which differential exposure misclassification by socioeconomic status position (SES) may exist, a companion study, described here, used personal monitoring of young children via their caregivers to 1) evaluate determinants of personal exposure for both poor and non-poor subjects selected from a population-based sample; 2) identify evidence of differential exposure misclassification by SES, and 3) assess the validity of ambient monitoring as a surrogate for personal exposures

2 Methods The study design was a repeated measures survey of subjects selected from a representative sample of households from the ex-tremes of the household income distribution within two of the 19 geographic districts in HCMC (seeFigs S1 and S2 in Supplemental Materialfor study design and sampling scheme) For each

24-h time activity diaries on 9 occasions spanning both the dry and rainy seasons between July 2007 and March 2008 Similar air quality measurements were made atfixed location monitoring sites

in each district Household characteristics were assessed by a questionnaire

2.1 Selection of households and participants Participating households were enrolled from a population-based sample of two districts within Ho Chi Minh City, the largest city in Vietnam and home to 6.1 million inhabitants (2004 Census)

In March 2007, The Bureau of Statistics conducted a 1000 house-hold survey (Fig S1 in Supplemental Material) to identify eligible households in Binh Thanh (BT) and District 2 (Fig 1) These districts are the closest to two key monitoring stations providing air quality data used in the HCMC hospital study (Le et al., 2012; Mehta et al.,

2013) To further increase the linkage with the hospital study, we selected households with young children (<5 years of age) BT is a densely populated district located in the city center, while District

2, located just across the Saigon river, is much less densely popu-lated, and during the time of the study could be considered somewhat peri-urban

Five wards were selected at random from within each of the two districts (Fig S2) From each ward, local officials provided a list of all

households per district were surveyed at random Information from this home survey (e.g expenditures, household size, assets) was used to assign households in each district to their corresponding expenditure quintiles and 16 households were selected at random

Supplemental Material) The primary caregivers for the young child in the household were selected for personal monitoring as they were likely to spend the greatest amount of time in close proximity to the young child, and thus most likely to experience a similar distribution of exposures

2.2 Analytical methods Between July 2007 and March 2008, nine repeated measure-ments of daily average personal exposures to PM2.5, PM10, and NO2 were made for each participant Participants were asked to wear a small (approximately 1.5 kg) backpack containing all sampling equipment while engaged in normal daily activities Participants were also trained on the proper removal and placement of

S Mehta et al / Atmospheric Environment 95 (2014) 571e580 572

backpacks during periods of long inactivity, such as during the

night, such that sampling inlets would remain as close as possible

to their breathing zones They completed a daily time activity diary

during each measurement period Detailed information on

expo-sure to potential sources of pollution, including traffic exposure,

incense, cottage industries, and tobacco smoke was recorded in

half-hourly intervals Participants also recorded whether or not

they were actually wearing the backpack during these intervals

Since participants were being monitored to represent exposures of

the young children under their care, they were also asked to

document the times when young children were with them during

the measurement period In addition, detailed information on

exposure to indoor sources of pollution, including incense and

mosquito coil use, tobacco smoke, proximity to traffic,

trans-portation mode and frequency, and ventilation quality in the house

was collected at the beginning of the sampling campaign by

interview with the primary caregiver Time-activity pattern (TAP)

diaries were completed at each of the home visits and the initial

household questionnaire wasfilled once with study technicians to

obtain information on demographics, self-reported information on

indoor pollutant sources and commuting (mode and time)

Household locations were measured by GPS (Garmin E-Trex Legend, Garmin International Inc., Olathe, KS) and the distance to the nearest monitoring site and nearest major road were calculated

in ArcGIS (v10, ESRI, Redlands, CA)

Personal PM10and PM2.5concentrations were measured for 24 h during each sampling session Leland Legacy (SKC) pumps were

PM2.5on 37-mm Teflon filters (No R2PJ037, Pall Life Sciences, Ann Arbor, MI) via PEM impactors (MSP Corporation, Shoreview, MN)

Model 510, BIOS, Butler, NJ) at the start and end of each measure-ment period to ensure consistentflow rates of 4.0 (±0.17) L/min Exposure to NO2was also monitored for each 24 h period using Ogawa passive samplers (Ogawa& Company USA, Pompano Beach FL)

Using the same methods as personal samples, dailyfixed loca-tion measurements were collected for the duraloca-tion of the eight

stations closest to the districts, i.e Zoo (closest to BT district) and District 2, to enable a comparison of personal exposures and ambient concentrations (Fig 1)

Fig 1 Map of HCMC, Vietnam showing monitoring fixed sites, participating homes, in Districts 2 and Binh Thanh (BT) and the road network (Open Map Street, Accessed Jan 15th 2013; ArcGIS v.10 ESRI).

All samples were analyzed at the HCMC Environmental

Pro-tection Agency (HEPA) exposure assessment laboratory The

labo-ratory included a temperature and humidity controlled glovebox

(Allen et al., 2001) and a microbalance for gravimetric analysis, a

reflectometer to measure particle absorbance, and an ion

chro-matograph to analyze the Ogawa samples Staff were fully trained

to carry out standardized analytical procedures

All filters were equilibrated in a glovebox in the exposure

assessment laboratory with controlled temperature (22.5± 2.5C)

and relative humidity (40± 5% RH) for 24 h prior to weighing with a

microbalance (Model SE2, Sartorius) on an anti-static weigh boat

Reflectance was measured, using a smoke-stain reflectometer (UK

Diffusion Systems Ltd., London, UK), and absorption coefficients

(ABS) were calculated according toISO 9835 standard (1993) All

ABS are reported in m1 105 Ogawa passive samplers were

assembled in HEPA's personal exposure assessment laboratory, and

kept refrigerated except during transport to and from the field

Aqueous extracts offilters were analyzed by ion chromatography

The average analyzed nitrite value from the extracts offield blanks

was subtracted from each sample extract's analyzed nitrite value

Subsequently, these blank corrected values were used to calculate

the concentrations in ppb

HEPAfield and lab staff carried out routine quality assurance

checks, including the collection of blank and duplicate samples,

balance stability testing in the laboratory, and use of blanks and

reference samples during laboratory analysis Specifically,

techni-cians used 1 duplicate for every 15filters, 1 laboratory blank for

every 10filters, 1 field blank per sampler, per week and one

co-located filter blank In total, 74 field blanks and 5 laboratory

and 5 laboratory blanks were deployed for thefixed site monitors

Duplicate NO2passive samplers were collected with the personal

(n¼ 120 based on 65 pairs of samples) and fixed site measurements

(n¼ 161 with 85 duplicates and 76 passive samplers) to assess

precision In each case, 4 laboratory blank samples were collected

PM concentration and absorbance values were excluded from

further analysis when meanflow was beyond ±5% of 4 L/min.; All

PM concentration and absorbance values were excluded when the

filter pre-weight was greater than the post-weight, or when the

sample duration was less than 20 h

All study participants signed informed consent forms prior to

their participation in the study In addition, at the end of each

monitoring period a honorarium of 100,000 VND (approximately

$7.00) was offered to each study participant This amount,

deter-mined by the local members of the collaboration, was intended to

compensate participants for their time and efforts without acting as

an unduefinancial incentive that could influence the poorer

par-ticipants' participation in the study The study proposal and

pro-tocols were reviewed and approved by the institutional review

board of the Biological and Medical Ethical Committee of HCMC

Department of Health (Decision no: 2751/SYT-NVY)

2.3 Statistical analysis

Correlations between monitors as well as between personal

measurements of NO2, PM10, PM2.5 and PM2.5absorbance were

examined in both pooled analysis and after aggregating over the

repeated measurements

All pollutant levels were examined in univariate analysis to

determine if any transformation of the data was required

Subse-quently, associations were evaluated between all air pollutant

measurements and distances to the closest monitor and road, and

with each variable from the time activity pattern (TAP) diary and

the initial household questionnaire For TAP variables, associations

were analyzed using mixed effects models with the participant as

random intercept and with unstructured covariance For the remaining variables, categorical variables were examined using either a t-test for binary predictors or ANOVA for categorical vari-ables with more than 2 categories

Generalized estimating equations were used to account for the correlated responses within each participant and we examined the relationship between personal and ambient concentrations in a sequential process by including as afixed effect: (Step 1) SES, (Step 2) District, (Step 3) both SES and District, (Step 4) Time Activity Pattern (TAP) initial household visit variables that were significant

in the bivariate analyses, (Step 5) SES, district, and all questionnaire variables In steps 1, and 3 to 5, the analysis was performed for each district where the participants reside, and also regardless of the location, using backward stepwise regression with a cut-off

p< 0.05 All analyses were conducted using Stata Version 10 (Sta-taCorp 2007 Stata Statistical Software: Release 10 College Station, TX: StataCorp LP)

3 Results 3.1 Descriptive results 3.1.1 Household questionnaire and time activity patterns (TAP) Participants wore the personal air sampler backpack for 16.5 h

them for an average of 17.4 h The average time a child was present with the woman who was surveyed was significantly higher (17.8 h)

in the poor households than in the non-poor homes (16.7 h) Overall, participants spent 93% of their time indoors The majority

of time spent in a household microenvironment was spent in the bedroom, followed by the living room

All households reported extensive use of fans, for an average of

15 h per day Incense use was also widespread, with 84% of the households reporting burning incense for an average burn time of

40 min per day 62/64 participants reported spending time in transit, of which 42% spent one hour or more in traffic each day While only 7 participants reported current smoking, 60 reported spending time in the presence of smokers and among those the average time of secondhand smoke exposure was 48 min per day There were few significant differences by district of residence and/

or SES in the time-activity patterns (Table 1) Subjects from non-poor households spent more time relaxing and used an air condi-tioner more frequently, while subjects from poor households spent more time smoking, in a room besides the kitchen, bedroom or living room, and used a fan more frequently Residents of District 2 spent more time in traffic, sleeping and in the bedroom, while residents of BT spent more time engaged in other activities, in the living room and using a fan

From the initial household questionnaire, only the use of mos-quito coils and the ventilation quality differed by district and by SES, while time spent in proximity to traffic when not commuting (e.g sitting or standing next to majors roads such as road-side stalls

or cafes) was significantly different by district and use of Kerosene

as cooking fuel differed by SES (Table 1)

3.1.2 Quality assurance results For NO2, 10%field blanks (n ¼ 74 for personal samplers and

sam-pling campaign with mean and standard deviation (sd) of 1 ppb (sd¼ 1.4 ppb) and 0.74 ppb (sd ¼ 0.7 ppb) leading to a limit of detection (LOD) of 5.2 and 2.8 ppb for personal andfixed sites, respectively 88% of the personal monitoring samples were above the LOD while only 75% of thefixed sites samples were above their corresponding LOD Duplicate samples were 13% and 21% of the total sample size for personal andfixed site samples, respectively

S Mehta et al / Atmospheric Environment 95 (2014) 571e580 574

There were no significant differences between paired samples and

a high Pearson correlation between paired samples was found for

thefield study pairs (r ¼ 0.8 for personal and r ¼ 0.6 for fixed

sites)

Of the 566 expected personal PM samples, 32 PM2.5and 36 PM10

values from the household measurement campaign were excluded

from the database prior to analysis due to either a large drift (i.e

±5% of 4 L/min) in the pump flow (n ¼ 17 for PM2.5and n¼ 20 for

PM10), missing pumpflow rates (n ¼ 11 for PM2.5and n¼ 11 for

PM10), invalidfilter weights including missing values and

pre/post-weight blankfilters that were too high to enable the calculation of

PM concentration (n¼ 11 for PM2.5and n¼ 11 for PM10) Similarly,

of the 86 and 103fixed-site samples collected at each of the District

2 and BT monitoring sites for each PM fraction size, there were 9

and 14 that were discarded in District 2 and BT, respectively for

PM10, and 10 and 13 for PM2.5in District 2 and BT, respectively, for

similar reasons (drift in pumpflow, missing flow rates, unusable

pre-post weights) as well as non-plausible PM concentration values

where PM2.5/PM10> 1 (n ¼ 2 for each of PM size fraction) For absorbance, 4 samples for each size fraction in thefixed site mea-surements were excluded due to negative values while 1 absor-bance measurement for each of PM2.5and PM10was eliminated in the personal samples

3.2 Pollutant levels After blank correction was applied, the mean ambient concen-tration of NO2in the BT District was statistically higher than that measured in District 2 (Table 2) The two monitors' measurements were moderately correlated (r¼ 0.48 p < 0.001) The mean personal concentrations of NO2were higher among the participants classi-fied as non-poor (21.5mg/m3, sd¼ 9.9) compared with those

clas-sified as poor (18.9mg/m3, sd¼ 10.6) (p ¼ 0.06) However, when examining these differences in personal concentrations by district,

we found that in the BT district this difference was in the expected (poor> non-poor) direction unlike in District 2

Table 1

Average time (standard deviation in hours per day) spent in microenvironments and on activities (TAP diaries), and initial household questionnaire descriptive statistics by district and SES.

Household activities

Microenvironment (home)

Microenvironment (outside)

Other personal factors

Household questionnaire

Do you burn incense?

Do you use mosquito coils?*x

Do you cook for sale outside?

Do you spend time close to traffic when not commuting?*

Ventilation quality in kitchen*x

Cooking fuelx

x Significant difference by SES (p < 0.05).

* Significant difference by District (p < 0.05).

Following a stratified (by district) analysis, there were no

sta-tistically significant differences in personal PM concentrations by

SES in BT district, but the difference was still significant in District 2

for both PM10and PM2.5(Table 3) As hypothesized, personal

con-centrations across districts for both PM sizes were significantly

higher among those classified as poor compared to non-poor

Ambient concentrations of both PM2.5and PM10were signi

fi-cantly higher in the BT district compared with district 2 However,

since the correlation between the 2 monitors was high and

statis-tically significant PM10: mean r¼ 0.8 p < 0.001 and PM2.5: mean

r¼ 0.9 p < 0.001, ambient PM levels were averaged across monitors

for the longitudinal comparisons with personal PM concentrations

Table 4below displays the levels of PM concentration and

absor-bance after averaging across the two monitoring sites

When examining the difference in personal and ambient

con-centrations by season (Fig 2), we found statistically significant

differences in PM2.5, PM10and absorbance for personal andfixed

site levels with higher levels in the dry compared with the rainy season For NO2, personal levels were slightly (p¼ 0.1) higher in the rainy season (18.8 ppb) compared with the dry season (17.2 ppb)

different by season, with higher levels in the rainy (23.1 ppb) vs dry (17.9 ppb) dry season

3.3 Correlations between outdoor and personal pollutants Overall, personal exposures were more highly correlated with concentrations measured at thefixed sites for particulate matter (median Spearman's r¼ 0.7 for both size fractions with BT monitor) compared with NO2(r¼ 0.42 for BT)

Regardless of the residential location and for all pollutants examined, the correlations between personal measurements and

those at the BTfixed site The difference was more pronounced for NO2(D2: r¼ 0.38) compared with particulate matter (D2: r ¼ 0.5 and 0.65 respectively for PM10and PM2.5)

correlated we examined the correlation between the average of the two monitoring stations with the repeated measurements of study participants.Fig 3shows the results of the analysis of the corre-lation by SES for absorbance, PM, and NO2with the latter being the correlation with the closest monitor to which a participant's home was located

Summary estimates (mean and median) of correlations showed much stronger differences by SES, with the correlations among the non-poor much better than those among the poor for all pollutants

Table 2

NO 2 summary statistics by district and by SES.

Overall Poor (n ¼ 104) Non-poor (n ¼ 112) Overall Poor (n ¼ 116) Non-poor (n ¼ 120)

Table 3

Personal and ambient PM concentration and absorbance levels by district and by SES.

Table 4

Mean fixed sites ambient PM concentration and absorbance.

PM concentrations (ug/m 3 )

PM ABS (m1 10 5 )

S Mehta et al / Atmospheric Environment 95 (2014) 571e580 576

except NO2(Fig 3) Results for PM2.5were amplified (PM10: r¼ 0.57

for non-poor vs r¼ 0.43 for poor participants; PM2.5: r¼ 0.62 for

non-poor vs r¼ 0.37 for poor participants) Differences by season

were also found only for PM2.5with higher correlation in the rainy

season compared with the dry season (Table 5)

Since we collected repeated measurements for each participant,

we alsofit a mixed effects model with subject as random intercept

to account for the correlation between visits (Table 6) The

modeling results showed the effect of place in the association

be-tween personal, ambient and SES

ambient and personal PM2.5and PM10concentrations regardless of

the ambient monitor used for comparison When examining our models separately in each district, we found that in District 2, being classified as poor (vs non-poor) explains significant additional variability in personal concentrations above what was explained by the ambient PM concentrations alone In BT however, SES did not

influence the association between ambient and personal PM2.5or

PM10concentration

Unlike PM, the associations between personal and ambient NO2 concentrations were not affected by the participant's SES, con-firming the results shown with the summary estimates of correlations

Finally, being exposed to air pollution in the rainy season or the dry season did not affect the association between ambient and personal concentrations in both districts

3.4 Exposure factors: determinants of personal concentrations

measurements, ambient PM was averaged between the 2 sites for all subsequent modeling In examining whether the association between ambient and personal concentrations was modified by SES

or other activities and/or time spent in different micro-environments and/or activities, we built determinants of personal

PM concentrations models (Table 7)

The determinants of exposure modeling indicated that SES and the time in which air conditioning (AC) was used both predicted the personal exposure for PM2.5and PM10in the expected direction (i.e stronger association for non-poor compared with poor participants and lower personal concentration with increased time of AC usage) For a standard deviation increase in ambient concentration of PM2.5 (21 mg/m3) and PM10 (38 mg/m3), the personal concentration increased by 18.5 and 57 mg/m3 respectively For a 120 min (1 standard deviation) increase in AC use, the personal PM2.5and PM10

Fig 2 Personal and ambient concentrations and absorbance by season for PM 2.5 , PM 2.5 absorbance, PM 10 and NO 2

Fig 3 Box plots of PM 2.5 and NO 2 individual longitudinal correlations between

per-sonal and ambient measurements.

concentration decreased by 1.4 and 5.5 mg/m3 respectively In

addition, smoking was a significant predictor of PM2.5exposures,

while distance to the nearest road (as provided by the initial

household questionnaire) was positively associated with the

per-sonal concentration of PM10, but not PM2.5nor absorbance Season,

a categorical variable relatively balanced among the two strata of

SES (31 poor subjects provided samples in each of the rainy and dry

season vs 35 and 33 for the non-poor study participants), had a

different effect on the personal level of PM2.5absorbance compared

with the personal levels of NO2 For the latter, being in the rainy

season increased the personal concentration of NO2 by 1.8 ppb,

implying a stronger association between ambient and personal

decreased by 0.62 m1105, for the rainy vs dry season leading to

a weaker outdoor to personal association in the rainy season

compared with the dry season

For NO2, both in District 2 and BT, questionnaire variables

explained more variability in personal concentration than the

so-cioeconomic position of the study participants The quality of the

ventilation in the kitchen was an important factor in the personal

concentration as every unit drop in ventilation quality (e.g from

moderate to bad) was associated with 2.5 and 2.3 ppb decrease in

the personal concentration in D2 and BT respectively which

cor-responds approximately to afive percentile downshift

Regarding modelfit, the determinants of personal PM

concen-tration for both PM size fractions explained less between-subject

variability compared with absorbance and NO2 It is important to

note however, that direct comparison of goodness offit for these

models is not feasible since the main predictors differed as a

function of the pollutant that was considered

4 Discussion

Using monitoring and modeling based approaches, we

evalu-ated whether poorer children in Ho Chi Minh City systematically

experienced higher exposures to air pollution per level of ambient

air pollution on any given day compared to non-poor children,

regardless of district of residence By comparing more precise es-timates of individual personal exposure with eses-timates based on the ambient monitoring stations, we were able to explore sys-tematic daily differences in exposuree major sources and levels e across socioeconomic position

We found that measured personal exposure was not well rep-resented by ambient concentration measurements in most cir-cumstances This is because exposure while partly reflected by

neighborhood“hot spots” as well as micro-environmental levels experienced by individuals according to their personal behaviors

We compared measurements of individual personal exposure with estimates based on concentrations measured at ambient moni-toring stations and found that there were systematic differences in these relationships across socioeconomic position and seasons for

were less correlated to those estimated from ambient monitors

In addition, ambient monitoring substantially underestimated personal exposures for all measured pollutants in Ho Chi Minh City, with a significantly higher underestimation among the poor for fine

PM Daily mean concentrations for PM measured at thefixed sites during the same time period were lower than the personal mea-surements, with BT district showing higher levels compared to those measured in District 2 (95.2 vs 77.8mg/m3for PM10and 50.1 vs.39.2mg/m3for PM2.5) Similar results were apparent for NO2with

sites, with significantly higher concentrations in BT district compared with District 2, and significant differences between poor and non-poor participants only in District 2

Thus, localized sources appeared to contribute to exposure error arising from the use of ambient monitoring site data for health effects assessments, Further, the relative contribution of different sources of exposure differed by socioeconomic position

A wide distribution of daily personal exposures to PM10

64.6mg/m3respectively, along with mean NO2personal exposure of 16.2 ppb This is consistent with the distribution of ambient air

Table 5

Summary estimates (mean and median) of individual longitudinal correlation between ambient (mean levels between two fixed site monitors for PM, and nearest monitor for

NO 2 ) and personal pollutants levels by SES and by season.

a Statistically different by SES.

b Statistically different by season.

Table 6

Effect of district and SES in personal/ambient concentrations repeated measures models.

Personal measurements

S Mehta et al / Atmospheric Environment 95 (2014) 571e580 578

pollution levels in HCMC, which are generally higher than those

reported in developed countries, but lower than levels observed in

other Asian mega-cities Personal concentrations for both PM sizes

were significantly higher among those classified as poor compared

to participants who were classified as non-poor Zhou and

col-leagues also demonstrated an SES gradient in PM levels in Accra,

Ghana (lowest PM in the high-SES neighborhood, and highest in

two of the low SES slums with geometric means reaching 71 and

131mg/m3forfine and coarse PM) (Zhou et al., 2011)

ambient monitors were 0.4 for NO2, 0.6 for PM2.5and PM10and 0.7

for absorbance These correlations were somewhat lower than

those observed in similar studies (Brunekreef et al., 2005; Janssen

et al., 1998, 2005; Noullett et al., 2006; Wallace, 2000) conducted

in developed countries (median longitudinal correlation (#

days)¼ 0.74 (4e8), 0.73 (10), 0.49 (2days for 23 weeks), for PM2.5,

PM10, Absorbance, and NO2, respectively)

Along with the socioeconomic gradient found in exposure to

PM in HCMC, the exposures of the non-poor were more highly

correlated with ambient measurements for both PM size fractions

while those found for NO2were not significantly affected by SES

This suggests that different PM sources may be influencing the

exposures of the poor and non-poor Our analysis of the household

characteristics and time activity patterns collected along with the

personal sampling campaign shed some light on these sources as

well as factors that would alter the relationship betweenfixed site

and personal measurements For instance, the quality of the

ventilation in the kitchen was significantly different between the

two SES strata, with the poor having worse ventilation quality

main predictors of the model for personal exposures From the

TAP diaries, differences in personal factors between the poor and

non-poor were more predominant than time spent in different

micro-environments as we observed statistically significant

dif-ferences between poor and non-poor HCMC residents

partici-pating in the study: the poor smoked and used fans more, while

the non-poor were more frequent users of AC In order, to

disen-tangle the roles of all the factors captured in the questionnaires

and the daily diaries from the role played by SES, we examined the

without including SES and offering all significant predictors in the

bivariate analysis; second forcing SES in the same models Should

the TAP and questionnaire variables be explained by the

socio-economic position, the multicollinearity would lead to only the

de-terminants of personal exposures

Overall, the models for the determinants of personal exposure to

NO2, PM10, PM2.5and absorbance indicated ventilation quality and

time spent in the kitchen, AC use and season as important factors

that were not fully captured by SES differences These results

indicate that epidemiologic analysis examining the effects of air pollution on health may be biased if surrogates of SES are not included Furthermore, more detailed information capturing the specificities of developing countries (e.g ventilation quality and AC use) would reduce the potential for different degrees of exposure misclassification that may be related to SES Other influential in-door air quality determinants, such as type of cooking devices used may have provided further insight in the SES gradient found in the examined pollutants; for although nearly all households (92%) used LPG as their cooking fuel, kerosene use was elevated in the poor (12.5%) compared to the non-poor (3%) households

Results of this study also aid in the interpretation of the com-panion hospital study, where analyses were not able to identify differential effects by socioeconomic position (Mehta et al., 2013)

In the hospital study, a single daily measurement of pollution was assigned to all children for a particular day As such, daily differ-ences in individual exposures across districts or socioeconomic groups could not be adequately assessed This study lends further support to the hypothesis that poorer children in Ho Chi Minh City systematically experience higher exposures to air pollution per unit

of reported ambient air quality on any given day compared to non-poor children, regardless of district of residence If the exposures of the poor are less well correlated with measurements made at the fixed sites used in epidemiologic analyses, there will be more exposure misclassification among the poor This would be expected

to result in a decreased ability to assess the true association be-tween short-term air pollution exposure and adverse health out-comes among the poor, and will limit the ability to assess differences in risk by socioeconomic position Our investigation is based on the premise that the siting of the two ambient monitors is representative of average ambient concentrations within the sur-rounding area where participants resided We examined and confirmed that (1) residents were living at similar distances to the nearest major road (245 m in BT vs 267 m in District 2 based on study technicians report), and (2) that road density was not significantly different around households and the corresponding monitor in each district However, we have no data to examine the distribution of industries across the two districts, although most industries are small-scales operations and located mainly within residential areas

Differential exposure to major sources of pollution, further

influenced by characteristics of Ho Chi Minh City's rapidly urban-izing landscape, resulted in systematically higher exposures among the poor Our experience documents potential for differential

ambient pollution monitors located in areas that differ in the relative contribution of different sources of pollution and other aspects of the urban environment correlated with SES These re-sults underscore the need to carefully evaluate how socioeconomic position may influence exposure to air pollution

Table 7

Final explanatory models showing significant variables affecting the association between personal and ambient NO 2 , PM 2.5 and PM 10 concentrations and absorbance.

AC: Air conditioning.

The authors would like to acknowledge the contributions of

HEPA field and lab staff, the International Scientific Oversight

Committee, and Timothy McAuley as well as the Bureau of Statistics

field staff

This project is supported with funds from the Health Effects

Institute and the Poverty Reduction Cooperation Fund of the Asian

Development Bank (Technical Assistance TA 4714-VIE), as well as

in-kind support from the Government of Vietnam

Appendix A Supplementary data

Supplementary data related to this article can be found athttp://

dx.doi.org/10.1016/j.atmosenv.2014.07.011

References

Allen, R., Box, M., Liu, L.-J.S., Larson, T.V., 2001 A cost-effective weighing chamber

for particulate matter filters J Air Waste Manag Assoc 51, 1650e1653 http://

dx.doi.org/10.1080/10473289.2001.10464392

Brunekreef, B., Janssen, N.A.H., de Hartog, J.J., Oldenwening, M., Meliefste, K.,

Hoek, G., et al., 2005 Personal, indoor, and outdoor exposures to PM2.5 and its

components for groups of cardiovascular patients in Amsterdam and Helsinki.

Res Rep Health Eff Inst., 1e70 discussion 71e79

Finkelstein, M.M., Jerrett, M., Sears, M.R., 2005 Environmental inequality and

cir-culatory disease mortality gradients J Epidemiol Community Health 59,

481e487 http://dx.doi.org/10.1136/jech.2004.026203

HEI International Scientific, 2010 Outdoor Air Pollution and Health in the

Devel-oping Countries of Asia: a Comprehensive Review Special report 18

ISO 9835, 1993 Ambient Air e Determination of a Black Smoke Index

Janssen, N.A., Hoek, G., Brunekreef, B., Harssema, H., Mensink, I., Zuidhof, A., 1998.

Personal sampling of particles in adults: relation among personal, indoor, and

outdoor air concentrations Am J Epidemiol 147, 537e547

Janssen, N.A.H., Lanki, T., Hoek, G., Vallius, M., de Hartog, J.J., Van Grieken, R., et al.,

2005 Associations between ambient, personal, and indoor exposure to fine

particulate matter constituents in Dutch and Finnish panels of cardiovascular

patients Occup Environ Med 62, 868e877 http://dx.doi.org/10.1136/

oem.2004.016618

Laurent, O., Bard, D., Filleul, L., Segala, C., 2007 Effect of socioeconomic status on the relationship between atmospheric pollution and mortality J Epidemiol Com-munity Health 61, 665e675 http://dx.doi.org/10.1136/jech.2006.053611

Le, T.G., Ngo, L., Mehta, S., Do, V.D., Thach, T.Q., Vu, X.D., et al., 2012 Effects of short-term exposure to air pollution on hospital admissions of young children for acute lower respiratory infections in Ho Chi Minh City, Vietnam Res Rep Health Eff Inst., 5e72 discussion 73e83

Lim, S.S., Vos, T., Flaxman, A.D., Danaei, G., Shibuya, K., Adair-Rohani, H., et al., 2012.

A comparative risk assessment of burden of disease and injury attributable to

67 risk factors and risk factor clusters in 21 regions, 1990e2010: a systematic analysis for the Global Burden of Disease Study 2010 Lancet 380, 2224e2260 http://dx.doi.org/10.1016/S0140-6736(12)61766-8

Mehta, S., Ngo, L.H., Dzung, D.V., Cohen, A., Thach, T.Q., Dan, V.X., et al., 2013 Air pollution and admissions for acute lower respiratory infections in young chil-dren of Ho Chi Minh City Air Qual Atmos Health 6 (1), 167e179 http:// dx.doi.org/10.1007/s11869-011-0158-z

Noullett, M., Jackson, P.L., Brauer, M., 2006 Winter measurements of children's personal exposure and ambient fine particle mass, sulphate and light absorbing components in a northern community Atmos Environ 40, 1971e1990 http:// dx.doi.org/10.1016/j.atmosenv.2005.11.038

Sarnat, J.A., Koutrakis, P., Suh, H.H., 2000 Assessing the relationship between per-sonal particulate and gaseous exposures of senior citizens living in Baltimore,

MD J Air Waste Manag Assoc 50, 1184e1198 Smith, K.R., Samet, J.M., Romieu, I., Bruce, N., 2000 Indoor air pollution in devel-oping countries and acute lower respiratory infections in children Thorax 55, 518e532

Wallace, L., 2000 Correlations of personal exposure to particles with outdoor air measurements: a review of recent studies Aerosol Sci Technol 32, 15e25 http://dx.doi.org/10.1080/027868200303894

Wang, H., Dwyer-Lindgren, L., Lofgren, K.T., Rajaratnam, J.K., Marcus, J.R., Levin-Rector, A., et al., 2012 Age-specific and sex-specific mortality in 187 countries, 1970e2010: a systematic analysis for the Global Burden of Disease Study 2010 Lancet 380, 2071e2094 http://dx.doi.org/10.1016/S0140-6736(12)61719-X Wong, C.-M., Vichit-Vadakan, N., Kan, H., Qian, Z., 2008 Public health and air pollution in Asia (PAPA): a multicity study of short-term effects of air pollution

on mortality Environ Health Perspect 116, 1195e1202 http://dx.doi.org/ 10.1289/ehp.11257

Wong, C.M., Vichit-Vadakan, N., Vajanapoom, N., Ostro, B., Thach, T.Q., Chau, P.Y.K.,

et al., 2010 Part 5 Public health and air pollution in Asia (PAPA): a combined analysis of four studies of air pollution and mortality Res Rep Health Eff Inst., 377e418

Zhou, Z., Dionisio, K.L., Arku, R.E., Quaye, A., Hughes, A.F., Vallarino, J., et al., 2011 Household and community poverty, biomass use, and air pollution in Accra, Ghana PNAS 108, 11028e11033 http://dx.doi.org/10.1073/pnas.1019183108

S Mehta et al / Atmospheric Environment 95 (2014) 571e580 580