Air pollution and admissions for acute lower respiratoryinfections in young children of Ho Chi Minh City

Sumi Mehta & Long H. Ngo & Do Van Dzung &

Aaron Cohen & T. Q. Thach & Vu Xuan Dan &

Nguyen Dinh Tuan & Le Truong Giang

Received: 8 April 2011 /Accepted: 16 August 2011 /Published online: 2 September 2011# Springer Science+Business Media B.V. 2011

Abstract This study assessed the effects of exposure to airpollution on hospitalization for acute lower respiratoryinfection (ALRI) among children under 5 years of age inHo Chi Minh City (HCMC) from 2003 to 2005. Case-crossover analyses with time-stratified selection of controlperiods were conducted using daily admissions for pneu-monia and bronchiolitis and daily, citywide averages of

PM10, NO2, SO2, and O3 (8-h maximum average) estimatedfrom the local air quality monitoring network. Increasedconcentrations of NO2 and SO2 were associated withincreased admissions in the dry season (November toApril), with excess risks of 8.50% (95%CI 0.80–16.79)and 5.85% (95%CI 0.44–11.55), respectively. PM10 couldalso be associated with increased admissions in the dryseason, but high correlation between PM10 and NO2 (0.78)limits our ability to distinguish between PM10 and NO2

effects. In the rainy season (May–October), negativeassociations between pollutants and admissions wereobserved. Results of this first study of the health effectsof air pollution in HCMC support the presence of anassociation between combustion-source pollution and in-creased ALRI admissions. ALRI admissions were generallypositively associated with ambient levels of PM10, NO2,and SO2 during the dry season, but not the rainy season.Negative results in the rainy season could be driven byresidual confounding present from May to October.Preliminary exploratory analyses suggested that seasonaldifferences in the prevalence of viral causes of ALRI couldbe driving the observed differences in effects by season.

Keywords Air pollution . ALRI . Children’s health .

Vietnam

AbbreviationALRI Acute lower respiratory infectionCI Confidence intervalCH1 Children’s Hospital Number 1CH2 Children’s Hospital Number 2HCMC Ho Chi Minh CityHEPA HCMC Environmental Protection AgencyICD-10 International Classification of Diseases, 10th

revision

Electronic supplementary material The online version of this article(doi:10.1007/s11869-011-0158-z) contains supplementary material,which is available to authorized users.

S. Mehta (*) :A. CohenHealth Effects Institute,101 Federal Street, Suite 500,Boston, MA 02110, USAe-mail: sumi.mehta@gmail.com

L. H. NgoBeth Israel Deaconess Medical Center and Harvard MedicalSchool,Boston, MA, USA

D. Van DzungHCMC University of Medicine & Pharmacy,Ho Chi Minh City, Vietnam

T. Q. ThachUniversity of Hong Kong,Pokfulam, Hong Kong

V. X. DanHCMC Center for Occupational and Environmental Health,Ho Chi Minh City, Vietnam

N. D. TuanHCMC Environmental Protection Agency,Ho Chi Minh City, Vietnam

L. T. GiangHCMC Department of Health,Ho Chi Minh City, Vietnam

Air Qual Atmos Health (2013) 6:167–179DOI 10.1007/s11869-011-0158-z

IMCI WHO/UNICEF Integrated Management ofChildhood Illness

NO2 Nitrogen dioxideO3 OzoneOR Odds ratioPM10 Particulate matter ≤10 μm in aerodynamic

diameterPM2.5 Particulate matter ≤2.5 μm in aerodynamic

diameterRSV Respiratory syncytial virusSD Standard deviationSO2 Sulfur dioxideWHO World Health Organization

Introduction

Background and significance

There is a growing body of epidemiological evidence thatexposure to particles generated by emissions from diversesources result in significant adverse health effects in urbanpopulations (Cohen et al. 2004). Children that live close toheavily trafficked roads experience greater adverse respira-tory episodes than children that live further away (Kim etal. 2008). In addition, a number of toxicological studieshave shown that exposure to particles from traffic emissionsresults in inflammatory responses in vitro and in vivostudies (HEI Panel on the Health Effects of Traffic-RelatedAir Pollution 2009).

In Asia, however, the composition of the emitted particlesdiffers considerably from North America and Europe wherethe majority of these studies have been performed. Vehiclefleets in Asia are dominated by two- and three-wheeledvehicles and automobile and truck fleets are significantlyolder (Han and Naeher 2006). In addition, there are a numberof local sources, which contribute significantly to exposures,but are not present in North America and Europe. Asignificant fraction of houses and roadside stalls rely onsolid fuels for cooking and heating, trash is frequentlyburned in the street, and a large fraction of the populationcontinues to use tobacco products in the home (HEIInternational Scientific Oversight Committee 2004).

With extensive numbers of the world’s population livingin highly polluted areas of Asia’s cities, increased effects ofair pollution on the health of these populations would havesignificant public health impact and highly relevant policyimplications. Although a recent systematic review revealed421 studies on the health effects of air pollution in Asia(Health Effects Institute 2006), to date, no studies havebeen conducted in many of the poorer Southeast Asiancountries, such as Laos, Cambodia, and Vietnam. The

ability to conduct such studies is currently compromised bythe relative lack of reliable and easily accessible data onhealth outcomes, routinely collected air quality data, andcollaboration between health and environment sectors.

Air pollution and acute lower respiratory infection

The capacity for combustion-derived air pollution to affectresistance to infection is well documented (Thomas andZelikoff 1999). More recent studies suggest a role for fineparticles (PM2.5) (Zelikoff et al. 2003). Effects on airwayresistance, epithelial permeability, and macrophage functionhave been associated with various components of thecomplex mixture of air pollution generated by indoor andoutdoor sources. There has also been considerable recentinterest in the role of particle-associated transition metals,including iron, in producing oxidative stress in the lung(Ghio 2004; Ghio and Cohen 2005), which has beenhypothesized to be a common factor in a range of adverseeffects of air pollution on the cardiovascular and respiratorysystems (Kelly 2003). PM-associated transition metals havealso been associated with altered host defenses in rats(Zelikoff et al. 2002).

Acute lower respiratory infections, including pneumonia,bronchitis, and bronchiolitis, are the largest single cause ofmortality among young children worldwide and thusaccount for a significant global burden of disease world-wide (Williams et al. 2002; World Health Organization2004). According to recent estimates, these infections causenearly one fifth of mortality in children under the age of5 years, with 90% of acute lower respiratory infection(ALRI) deaths being directly attributable to pneumonia(World Health Organization 2004). A substantial fraction ofthe burden is experienced by populations in Asia andAfrica; the annual incidence of lower respiratory infectionsis 134 million in Asia and 131 million in Africa out of anoverall global annual total of 429.2 million cases for allages. For example, and of relevance to this investigation,more than 33,000 ALRI deaths occur in Vietnam each year(World Health Organization 2004).

While outdoor air pollution has been associated withincreased ALRI morbidity and mortality, very few studieshave been conducted in developing countries of Asia,where populations are exposed to much higher levels of airpollution, and experience the greatest burden of disease dueto ALRI. Of the 42 studies reviewed by Smith et al. (2000)and Romieu et al. (2002), only three were conducted indeveloping countries of Asia, although the highest expo-sures and the greatest burden of disease due to indoor andoutdoor air pollution is borne by the populations in theregion (Cohen et al. 2004; HEI International ScientificOversight Committee 2004; Smith et al. 2004). As such, theresults of this study have the potential to make an important

168 Air Qual Atmos Health (2013) 6:167–179

contribution to the growing literature on the health effectsof air pollution in Asia.

Location

HCMC, formerly known as Saigon, is a major industrialand commercial center of Vietnam, and home to over sixmillion people. Rapid economic development continues tobring more migrants to the city, contributing to the trafficcongestion and urban crowding. Major sources of airpollution in HCMC include transport, energy, and industry.It should be noted that the vast majority of transportationdemand is met by motorbikes and/or motorcycles (56%)(Viet Nam Register 2002). Other key local sources of airpollution with the potential to result in large proportions ofexposure include smoking, the use of solid fuels, such aswood and coal, for cooking (particularly at roadside foodstalls), and incense burning.

The city has hot and humid weather year round, withmean temperatures averaging between 23°C and 32°C(73.4° to 89.6°F). There are two seasons in HCMC—adry season, from November to April, and a rainy season,from May to October. With daily average PM10 levelsroutinely ranging from 30 to over 150 μg/m3, HCMCprovides a unique opportunity to evaluate the health effectsof short-term changes in air pollution in a tropical climateand across a wide range of the exposure response curve.

Specific aim

Using routinely collected data on air quality and hospitaladmissions from January 1, 2003 to December 31, 2005,assess whether increased short-term (on the order of days)exposures to air pollution are associated with increasedfrequency of hospitalizations for acute lower respiratoryinfections among children under 5 years of age.

& Admissions for ALRI, specifically pneumonia andbronchiolitis, were extracted from computerized recordsof the two Children’s Hospitals of HCMC. Nearly allchildren admitted for respiratory illnesses in HCMC arehospitalized in one of the two pediatric hospitals. Thus,we captured nearly all children’s admissions forrespiratory illness in HCMC.

& Daily, city-level exposure estimates of particulate matterwith diameter less than 10 μm (PM10), O3, NO2, andSO2 were generated using data from the HCMCEnvironmental Protection Agency’s (HEPA) ambientair-quality monitoring network.

& Daily meteorological information including temperature,relative humidity, and rainfall were collected from KTTVNB, the Southern regional hydro-meteorological center.

Methods

This study was approved by the institutional reviewboard of the Biological and Medical Ethical Committeeof HCMC Department of Health (Decision no: 2751/SYT-NVY).

Patient population

We focused the study inference on the children of HCMCunder 5 years of age. HCMC’s two pediatric hospitals,Children’s Hospital Number 1 (CH1) and Children’sHospital Number 2 (CH2), cover nearly all pediatricadmissions in the city. CH1, located in District 10, is a900-bed hospital and had 1,071,756 outpatient visits and48,854 admissions in 2004. Located in District 1, the 800-bed CH2 had 587,718 outpatient visits and 54,629 hospitaladmissions in 2004. Diseases of the respiratory system are aleading cause of inpatient admission in both hospitals.

ALRI is diagnosed on the basis of WHO’s IntegratedManagement of Childhood Illness (IMCI) criteria. Otherthan severity of illness, we have not been able to identifyany other major factors that could affect the likelihood of achild’s admission.

& Hospital capacity: the hospitals adhere to the WHO/UNICEF IMCI guidelines for the management of ALRIwhen determining the need for admission. Despiteincreasing numbers of beds over time, both hospitalsoperate at or beyond capacity on a regular basis.Clinicians have reassured the investigative team thatadmissions for ALRI are not impacted by bed capacity;when necessary, multiple patients share the samehospital bed.

& Family financial status: as ability to pay is determinedafter admission, family financial status does notinfluence admission.

& Location of residence: while a child who lives in aremote province (outside HCMC) may occasionally beadmitted using less stringent criteria, due to thehardship of traveling back and forth, since we focusonly on HCMC residents, this will not affect our study.

Respiratory outcome data

Ideally, we would have liked to collect information onALRI incidence. As only daily aggregate information onoutpatient visits is available, however, we focused ouranalysis on hospital admissions. Using incidence of hospitaladmissions not only allowed us to identify ALRI casessevere enough to warrant hospitalization but also limitedour ability to precisely ascertain the time when clinicallyrelevant disease onset occurred.

Air Qual Atmos Health (2013) 6:167–179 169

Criteria for ALRI diagnosis and admissions

The IMCI was launched in 1995 by WHO and UNICEF, inorder to address five leading causes of childhood deaths inthe world: pneumonia, diarrhea, measles, malaria, andmalnutrition. The Initiative has three main components:improvements in case-management skills of health staff;improvements in health systems; and improvements offamily and community practices (WHO-CAH 1997). Allpatients seen at the pediatric hospitals are diagnosed andadmitted using standardized IMCI criteria for diagnosingacute respiratory illness. Both hospitals use ICD-10 codesfor disease reporting and classification.

Admissions data collection and management

As nearly all children admitted for respiratory illnesses inHCMC are hospitalized in one of the two pediatrichospitals, we were able to capture nearly all children’sadmissions for respiratory illness in HCMC.

Local collaborators at the children’s hospitals informedus that they did not use objective clinical criteria fordistinguishing between pneumonia and bronchiolitis. Thus,we created a single outcome category for ALRI, whichincludes both pneumonia and bronchiolitis. ALRI admis-sions in children 5 years of age and under from January2003 to December 2005 were extracted from computerizedrecords of Children’s Hospitals 1 and 2 (CH1 and CH2)using the following criteria:

1. Admission date from January 01, 2003 to December31, 2005

2. Age at admission date less than 5 years3. Residence of HCMC on admission date: patients

residing in the five rural districts of HCMC wereexcluded, as their exposures are not well reflected bythe air quality monitoring network

4. Discharge diagnosis includes primary diagnosis ofICD-10: J13 to J18, or J21

5. Neonatal admissions (<28 days) were excluded, sincethese are likely to be influenced by perinatal conditions.

6. Consistent with other studies of ALRI in youngchildren (http://ehs.sph.berkeley.edu/guat/page.asp?id=32), we excluded all repeated visits occurring withinthe same 14-day period to avoid double counting of thesame case. This was only possible for CH1, due tolimitations in the electronic dataset for CH2. As only ahandful of cases were removed in CH1, however, thisis unlikely to have major implications for our studyresults.

7. Aside from this 14-day window restriction, all multiplevisits for children within the study period wereretrieved.

A quality assurance unit conducted manual cross checksto guarantee the quality of electronic data in the hospitaldatabases.

We chose not to include a control disease in our analysis,as we were not convinced that we would be able to selectan ideal “control” disease. Moreover, hospitals are operat-ing beyond capacity; while we have been assured that thisdoes not affect admission for lower respiratory illness, weare unclear how limited capacity may impact admission forother conditions.

Definition of case period

Since we focused on hospital admissions, we used anempirical induction time (Rothman and Greenland 1998), i.e. the definition of the case period took induction times forALRI, as well as time between onset of illness anddetection of disease/time of hospitalization into consider-ation. The induction time would likely be on the order of afew days. On the basis of information from primary healthclinics, hard copy clinical records, and physicians at theChildren’s hospitals, we assumed that the case periodshould be between 1 and 6 days. This has implications forthe choice of pollutant lag times; while we explored singleday lags from lag 0 (pollution on same day of admission) tolag 10 (pollution 10 days prior to admission), we focus onthe average lag (1–6) days, which incorporates pollutionlevels 1 to 6 days prior to admission.

Environmental data

Air quality data

The HEPA, with the technical assistance from the NorwegianInstitute for Air Research (NILU), has maintained nineautomatic air-quality monitoring stations to monitor PM10,O3, NO2, and SO2 around the city since 2001. Data from thefour background and/or residential stations were consideredpotentially eligible for inclusion in our analyses. Dailyaverage values were created for each monitoring station bytaking the mean of 24 hourly values for PM10, NO2, SO2,and by generating maximum 8 h moving averages for O3. A75% completeness criterion was applied to all hourly data.No additional constraints to the data, i.e. no thresholds, wereapplied.

Time-series plots and inter-station correlations werecarefully assessed to inform decisions about the qualityand completeness of station-specific average daily pollutantconcentrations and to assess spatial homogeneity. Hourlyvalues for each monitoring site were manually reviewed toflag recurrent values. All strings of four or more repeatedvalues, indicating a problem with the monitoring system,were deleted. Site-specific daily time-series were also

170 Air Qual Atmos Health (2013) 6:167–179

reviewed to identify other potential data quality concerns.Citywide estimates for each pollutant were created bytaking the mean of daily average data from each eligiblestation.

Meteorological data

Mean daily data on rainfall, temperature, and humiditywere calculated from hourly data provided by KTTV NB,the Southern regional hydro-meteorological center, andhave very few missing values.

Statistical methods

Analyses were conducted using case-crossover, time-stratified design. While the majority of earlier case-crossover design-based studies have used a symmetric bi-directional design, recent publications emphasize theadvantages of using a time-stratified design (D'Ippoliti etal. 2003; Janes et al. 2005a; Levy et al. 2001; Lumley andLevy 2000; Mittleman 2005). Specifically, the time-stratified design has been shown to produce unbiasedestimates, as it enables the removal of overlap bias resultingfrom time trends in the pollutant data. (Janes et al. 2005b).

Control days were every 7th days from the inductiondate within the same month as admission. Childrenhospitalized on the same day share the same case andcontrol periods. Data were analyzed using conditionallogistic regression where each subject has once case periodand a variable (depending on when during the month theadmission occurred) number of control periods. This isequivalent to a 1 M case control study. We assume that eachchild is hospitalized only once or that the repeated visitsoccurred more than 14 days apart and can be treated asindependent ALRI episodes.

As the time between onset of illness and hospital admissionwas thought to range from 1 to 6 days, it was not possible tospecify a priori a single day lag. On the basis of “typical”referral patterns and pathways to hospital admission, therelevant window of exposure was thought to be within theweek prior to hospital admission, i.e. 1 to 6 days before thedate of admission.We assessed results for single day lags fromlag 0 to lag 10, but emphasized results which used the average1- to 6-day lag, since this best reflects the case period. Theseresults take pollution levels in the 1 to 6 days leading up toadmission into account. All results were calculated using lag 0for temperature; the sensitivity of results to this assumptionwas explored in sensitivity analyses.

Before estimating the effect of PM10 controlling for theother gaseous pollutants, we examined the correlationamong PM10 and the other pollutants, since collinearitycould introduce an estimation problem for getting theadjusted estimate of PM10. All analyses were stratified by

season due to the observed difference in the distribution ofthe pollutants and/or meteorology between the dry andrainy seasons, and because ALRI incidence is known todepend on season as well.

Results

Hospital admissions

Table 1 summarizes the characteristics of admitted patients.There were a total of 15,717 admissions with 10,468 ofthese admissions occurring at CH2, nearly twice thenumber of admissions for CH1 (5,249). CH2 consistentlyhas more than twice the number of admissions than CH1.Admissions tended to peak between July and August eachyear, during the rainy season in HCMC (Fig. 1). Over 64%of admissions were male children, and around 74% ofchildren were under 2 years of age. ICD classification atdischarge was relatively consistent across hospitals; around58% of the ALRI admissions were discharged with thediagnosis of “pneumonia” and around 42% were dischargedwith a diagnosis of “bronchiolitis.”

Environmental data

Table 2 and Fig. 2a–d show the annual distribution ofcitywide pollutant concentrations across all monitoring sitesfor the duration of the study. The seasonal trend in the data,corresponding to the rainy and dry seasons in HCMC isevident, with pollutant levels at their highest in the dryseason and lowest in the rainy season. Concentrations of allpollutants, particularly PM10, show greater variability in thedry season.

With the exception of the correlation between PM10 andNO2 in the dry season, none of the correlations exceeded0.7 (Table 3). During the dry season, inter-pollutantcorrelations were highest for PM10 and NO2 (r=0.78) andPM10 and O3 (r=0.66), and lowest for PM10 and SO2

(0.32). During the rainy season, inter-pollutant correlationswere lower, with highest correlations observed betweenPM10 and O3 (r=0.60), and PM10 and SO2 (0.36), butlimited correlations observed among other pollutants.

Meteorological data

The average daily temperature in HCMC ranges from23°C to 32°C, and the average daily relative humidity isconsistently high, ranging from 51% to 93% (Table 2).The seasonal definition used is consistent with the rainfallexperienced during the study period, i.e. the observedseasonal variation corresponds to the dry and rainyseasons.

Air Qual Atmos Health (2013) 6:167–179 171

Statistical analyses

All excess relative risk estimates and confidence intervals arereported per 10 μg/m3 increase in pollutant concentrations.

Main results

Large seasonal differences in admission patterns andpollution levels were observed. Sixty percent of ALRIadmissions occurred during the rainy season, while thehighest pollutant concentrations were observed in thedry season. We initially adjusted for season to controlfor seasonal differences in admissions and pollutionlevels and checked for seasonal interaction in the single-pollutant Poisson Regression models. The significanceof the seasonal interaction term for NO2 (p<0.0008)provided us with further indication that we shouldconduct stratified analyses to reduce the potential forconfounding by season. We also conducted sensitivityanalyses to assess the adequacy of our binary classifica-tion of season.

Overall and season-specific results for single-pollutantand two-pollutant models are summarized in Tables 3 and 4and Fig. 3. Results differed markedly when analyses werestratified by (rather than simply adjusted for) season. ALRIadmissions were generally positively associated withambient levels of PM10, NO2, and SO2 during the dryseason, but not the rainy season.

Dry season results

Exposure to PM10 was weakly associated with a 1.25%(95% CI −0.55–3.09) excess risk of ALRI admissions forevery 10 μg/m3 increase in exposure. This risk remainedsimilar after adjusting for SO2 and O3. This association wasno longer observed after adjusting for NO2. Associationsbetween exposure to O3 and ALRI admissions were notobserved in the single pollutant models. O3 exposure alsoremained unassociated with any change in risk afteradjustment for other pollutants in the two-pollutant models.There was strong evidence of an NO2 effect, with excessrisk estimates ranging from 7% to 18% for every 10 μg/m3

Table 1 Selected characteristicsof ALRI admissions, 2003–2005 Children’s hospital 1 Children’s hospital 2 Combined

Sex

Male 71.7% 3761 61.4% 6424 64.8% 10185

Female 28.3% 1488 38.6% 4044 35.2% 5532

Age

0 to 1 years of age 38.4% 2015 29.6% 3101 32.6% 5116

1 to 2 years of age 41.0% 2150 40.5% 4243 40.7% 6393

2 to 5 years of age 20.7% 1084 29.8% 3124 26.8% 4208

Month

January 6.3% 332 6.4% 667 6.4% 999

February 0.0% 265 5.0% 520 5.0% 785

March 7.3% 385 6.2% 648 6.6% 1033

April 7.0% 370 6.3% 664 6.6% 1034

May 7.9% 417 8.0% 835 8.0% 1252

June 10.3% 541 10.2% 1064 10.2% 1605

July 11.5% 602 12.4% 1296 12.1% 1898

August 12.5% 654 11.2% 1173 11.6% 1827

September 9.1% 476 8.7% 914 8.8% 1390

October 9.0% 474 9.3% 975 9.2% 1449

November 6.9% 360 8.5% 887 7.9% 1247

December 7.1% 373 7.9% 825 7.6% 1198

Season

Rainy (May–October) 60.3% 3164 59.8% 6257 59.9% 9421

Dry (November–April) 39.7% 2085 40.2% 4211 40.1% 6269

Year

2003 35.0% 1836 34.9% 3654 34.9% 5490

2004 29.9% 1567 29.5% 3086 29.6% 4653

2005 35.2% 1846 35.6% 3728 35.5% 5574

172 Air Qual Atmos Health (2013) 6:167–179

increase in exposure in the single-pollutant models (lag 1 tolag 10). The highest increase in risk of ALRI admissionswas associated with exposure to NO2, with an excess riskof 8.50% (95% CI 0.80–16.79) for every 10 μg/m3 increasein exposure observed in the single-pollutant model. Theseeffects were robust to adjustment for other pollutants, andbecame even more pronounced after adjustment for PM10.There was limited evidence of a SO2 effect, with excess riskranging from 2% to 6% for every 10 μg/m3 increase inexposure. SO2 was associated with an increased risk ofALRI admissions in the single-pollutant model, with anexcess risk of 5.85 (95% CI 0.44–11.55) for every 10 μg/m3 increase in exposure. This association was relativelyrobust to adjustment for PM10 and O3, but the two-pollutantadjusted for NO2 indicated confounding of effects by NO2.

The magnitude of the effects observed for NO2 and SO2,along with the wide confidence intervals, can be partiallyexplained by the fact that excess risks were calculated forevery 10 μg/m3 increase in pollutant concentrations, whichis close to the standard deviation for these pollutants duringthe study period.

Rainy season results

Negative associations between PM10 and ALRI admissionswere observed in the rainy season. No association with O3

exposure was observed in the single pollutant models, butO3 was negatively associated with ALRI admissions in thetwo-pollutant models. There was little evidence of anassociation between NO2 and ALRI admissions in the

Fig. 1 Daily ALRI admissions to HCMC children’s hospitals, 2003–2005

Table 2 Distribution of mean daily air pollutants and meteorologic variables (citywide estimate), 2003–2005

Overall Dry season Rainy season

N Mean SD Min Max N Mean SD Min Max N Mean SD Min Max

PM10 (μg/m3) 1040 73.19 29 19 196 511 83.64 31 32.16 196 529 63.1 22.61 19 185.4

O3 (μg/m3) 1057 75.03 30 17 185 516 91.84 27 22.83 185 541 59 23.44 17 143.2

NO2 (μg/m3) 1022 22.1 7.7 5 55 498 23.06 8.1 8.4 50.6 524 21.2 7.14 5 55.17

SO2 (μg/m3) 720 21.58 11 2.7 80 416 26.37 11 5.95 80.4 304 15 7.59 2.7 37.45

Temperature (Celsius) 1096 28.19 1.41 23.10 32.00 544 28.1 1.5 23.1 31.7 551 28.3 1.3 24.7 32

Relative Humidity (%) 1096 73.71 7.49 51.10 93.40 544 69.3 6.3 51 88 552 78 5.9 60 93.4

Rainfall (cm) 1096 0.48 1.23 0.00 11.44 544 1.1 5.2 0 64.4 552 8.5 15.7 0 114.4

Air Qual Atmos Health (2013) 6:167–179 173

rainy season. The single pollutant estimate suggested anegative association between NO2 and ALRI admissions,but this effect was no longer apparent after adjustment forother pollutants. Associations between SO2 and ALRIadmissions were not observed in the rainy season.

Sensitivity analyses

Sensitivity analyses were conducted to examine thepotential impact of individual monitoring stations, temper-ature lag choice, exploration of seasonal effects, and

Fig. 2 a–d Daily average pollutant concentrations (μg/m3), citywide estimate, 2003–2005

Table 3 Interpollutant correla-tions, by season, 2003–2005 Dry season Rainy season

PM10 O3 NO2 SO2 PM10 O3 NO2 SO2

PM10 (μg/m3) 1.00 0.66 0.78 0.32 1.00 0.60 0.18 0.36

O3 (μg/m3) 0.66 1.00 0.44 0.19 0.60 1.00 0.17 0.65

NO2 (μg/m3) 0.78 0.44 1.00 0.29 0.18 0.17 1.00 0.01

SO2 (μg/m3) 0.32 0.19 0.29 1.00 0.36 0.65 0.01 1.00

174 Air Qual Atmos Health (2013) 6:167–179

monitor-specific effects on observed results. Using differenttemperature lags (0 versus average lag 1–6), includingrainfall as a continuous variable, and seasonal reclassifica-tion of selected time periods had little impact on results.

Discussion

Main results

Exposure to air pollutants was generally positively associ-ated with air pollution during the dry season (November–April) and inversely associated with pollution in therainy season (May–October). Results suggest that in-creased concentrations of NO2, SO2, and PM10 areassociated with increased hospital admissions for ALRIin young children of HCMC in the dry season, althoughSO2 and NO2 display the most robust relationships. Aswill be discussed later in more detail, PM10 could also beassociated with increased hospital admissions in the dryseason, but high correlation between PM10 and NO2 (0.78)limits our ability to distinguish between PM10 and NO2

effects.We know of no reason to think that exposure to air

pollution could reduce the risk of ALRI in the rainy season,and infer that these results could be driven by residualconfounding or other bias present within the rainy season.Although we could not specifically identify these sources ofbias, factors influencing the results in the rainy season could

potentially impact results observed in the dry season as well.The prevalence of respiratory illness is higher during the rainyseason, when there are likely to be other risk factors whichplay a stronger role than pollution levels. Pollutant levels areat their lowest during the rainy season. This situation increasesthe potential for negative confounding within the rainy season.

The results appeared robust to definition of season,inclusion of rainfall as a continuous variable, and otherpotential sources of error. No clear evidence of monitor-specific effects was observed; differences across monitoringstations had widely overlapping confidence intervals.

Comparison with other studies

Variation in disease classifications, averaging times, andseasonal definitions limits the ability to make directcomparisons with studies conducted elsewhere. Studies thathave found the strongest PM effects have focused specif-ically on bronchiolitis (Karr et al. 2006; Segala et al. 2008),while most of the studies, which have used a more broadlygrouped disease classification, have found similarly incon-clusive results for PM10 (Barnett et al. 2005; Gouveia andFletcher 2000; Hernandez-Cadena et al. 2007). Recentpublications that focus on the effects of sub-chronicexposures to air pollution have used averaging times onthe order of weeks or months and seem to suggest strongereffects (Karr et al. 2009). Similar to other studies focusedon the effects of acute effects; however, this study usedaveraging times on the order of days.

Table 4 Excess risk (ER%) per 10 μg/m3 increase in pollutant concentrations, overall and by season, average lag (1–6) days, single, andbipollutant models

Overall Dry Rainy

ER% 95% CI lo 95% CI hi ER% 95% CI lo 95% CI hi ER% 95% CI lo 95% CI hi

PM10 (μg/m3) Single −1.10 −2.31 0.12 1.25 −0.55 3.09 −3.11 −4.76 −1.42

Adj SO2 −0.57 −2.08 0.95 1.88 −0.15 3.95 −3.62 −5.90 −1.28Adj O3 −0.19 −1.60 1.25 2.03 −0.01 4.11 −2.18 −4.14 −0.19Adj NO2 −1.20 −2.60 0.22 −0.36 −3.02 2.37 −2.90 −4.67 −1.10

O3 (μg/m3) Single −1.96 −3.25 −0.64 −0.79 −2.67 1.13 −2.98 −4.78 −1.14

Adj SO2 −1.18 −2.75 0.42 −0.63 −2.78 1.56 −1.01 −3.51 1.56

Adj PM10 −1.98 −3.48 −0.45 −1.78 −3.87 0.36 −1.96 −4.13 0.25

Adj NO2 −2.07 −3.46 −0.67 −1.28 −3.27 0.74 −2.91 −4.85 −0.92NO2 (μg/m

3) Single −1.08 −5.14 3.17 8.50 0.80 16.79 −5.15 −9.94 −0.10Adj SO2 3.40 −2.39 9.53 12.07 2.76 22.22 −2.71 −9.98 5.16

Adj PM10 0.95 −3.81 5.94 9.70 −1.80 22.55 −2.42 −7.61 3.07

Adj O3 1.11 −3.30 5.72 10.12 1.93 18.97 −2.48 −7.73 3.06

SO2 (μg/m3) Single 2.61 −1.49 6.87 5.85 0.44 11.55 −2.13 −8.25 4.41

Adj PM10 2.77 −1.35 7.06 5.44 0.01 11.15 −0.81 −7.05 5.86

Adj O3 2.95 −1.18 7.25 5.72 0.31 11.43 −1.16 −7.76 5.92

Adj NO2 1.84 −2.31 6.16 3.70 −1.76 9.47 −1.78 −7.99 4.85

Air Qual Atmos Health (2013) 6:167–179 175

Although other studies have looked for potentialeffect modification by season, the definition of seasonused has varied from study to study. Seasonal differ-ences in effects are location-dependent, in that differentseasonal patterns translate into different trends intemperature, precipitation, and disease incidence. Studies

conducted in North America and Western Europe havefound effects in the “winter” season, where “winter”corresponds to the season where both pollution levelsand disease incidence are likely to be high (Karr et al.2006; Segala et al. 2008). A study conducted in Australiaand New Zealand (Barnett et al. 2005), with different

-10

-50

510

1520

-10

-50

510

1520

-10

-50

510

1520

sing

le

adj

SO

2

adj

O3

adj

NO

2

sing

le

adj

SO

2

adj

PM

10

adj

NO

2

sing

le

adj

SO

2

adj

PM

10

adj

O3

sing

le

adj

PM

10

adj

O3

adj

NO

2

SO2N2OO3PM10

SO2N2OO3PM10

SO2N2OO3PM10

sing

le

adj

SO

2

adj

O3

adj

NO

2

sing

le

adj

SO

2

adj

PM

10

adj

NO

2

sing

le

adj

SO

2

adj

PM

10

adj

O3

sing

le

adj

PM

10

adj

O3

adj

NO

2

sing

le

adj

SO

2

adj

O3

adj

NO

2

sing

le

adj

SO

2

adj

PM

10

adj

NO

2

sing

le

adj

SO

2

adj

PM

10

adj

O3

sing

le

adj

PM

10

adj

O3

adj

NO

2

3

a

b

c

Fig. 3 a–c Excess risk (ER%)per 10 μg/m3 increase in pollut-ant concentrations, overall andby season, average lag (1–6)days, single and bipollutantresults

176 Air Qual Atmos Health (2013) 6:167–179

seasonal patterns of temperature and disease foundstronger effects in the warm season.

To the best of our knowledge, this is the first study tofocus specifically on assessing differences by wet and dryseasons. HCMC has a tropical climate with little variationin temperature but distinct seasonal patterns with regard torainfall that are correlated with distributions of respiratoryinfections. Thus, while the seasonal definition used heremay not be directly comparable with other studies, it isappropriate for this particular study.

The sensitivity of PM10 results in the rainy season to theinclusion of data on respiratory syncytial virus (RSV)prevalence was explored using methods developed to assesswhether unmeasured confounders in observational studiescould cause a bias large enough to reverse estimates ofeffect (Lin et al. 1998). Preliminary results suggest thatseasonal differences in the prevalence of viral causes ofALRI could be driving the observed differences in effectsby season. For more details, see the Online Reference. IfALRI of bacterial etiology is more strongly associated withair pollution exposures than ALRI caused by viruses suchas RSV, it is possible that our models, which did not takeRSV into account, may have induced a spurious negativeassociation between ALRI and air pollution. In addition,with virtually no RSV incidence in the dry season, thesefindings also lend credibility to the notion that RSV couldinfluence results primarily in the rainy season.

Differentiating PM10 vs. NO2 effects

While effects in this study appear to be driven by exposure toNO2, the high correlation between PM10 and NO2 (0.78)limits our ability to clearly distinguish PM10 and NO2 effects.Indeed, other studies focused on the association between airpollution and ALRI in young children have noted thechallenges of adequately distinguishing between PM10 andNO2 effects (Gouveia and Fletcher 2000; Barnett et al. 2005;Braga et al. 2001). One study (Saldiva et al. 1994) that founda strong positive association between NOx and childrespiratory mortality, while PM10 effects were not observed.

PM10 is a complex mixture of components that, likeNO2, serves as an imperfect indicator of combustion-relatedpollutants, but also represents non-combustion/crustal sour-ces of pollution that are prominent in HCMC, such asconstruction. The much lower correlation between NO2 andPM10 during the rainy season provides further evidence thatthese indicator pollutants may not be accurately character-izing exposures to air pollution from combustion processesin the rainy season. PM2.5 data, which would serve as abetter indicator of combustion processes, are not routinelyavailable in HCMC, and differences in PM composition byseason also remain unknown. Nevertheless, taken as awhole, results suggest that increased risks of ALRI

admissions in young children are associated with increasesin combustion-related pollution (including, but not exclu-sive to traffic pollution).

A recent systematic review of the literature found thateach 10 μg/m3 increase in long term ambient PM2.5

concentrations is associated with around a 12% increasedrisk of ALRI incidence, with the results of short-termstudies reviewed indicating that the health effects of airpollution continue to be observed at higher concentrations,and across a range of geographic locations (Mehta et al.2011). In addition, exposure to air pollution from indoorcombustion of solid fuels has also been consistentlyassociated with increased incidence and mortality risk in14 studies in developing countries (Smith et al. 2004), andchildren’s exposure to second-hand smoke, defined ashaving one or both parents who smoke indoors, has alsobeen associated with increased incidence of ALRI infec-tions and hospital admissions (U.S. Department of Healthand Human Services 2006). Thus, while the results of thisstudy would be difficult to interpret in the absence of otherevidence, given the strong effect modification by seasonobserved, results support the presence of an associationbetween combustion-source pollution and increased ALRIadmissions in HCMC which is consistent with studiesconducted elsewhere.

This study is, to the best of our knowledge, the firststudy of ALRI admissions in young children to beconducted in an Asian city, and the first study of the healtheffects of air pollution to be conducted in Ho Chi MinhCity, Vietnam. Ambient pollution levels in HCMC arecertainly much higher than those experienced in theexisting air pollution and ALRI morbidity literature, butremain somewhat lower than levels in other Asian mega-cities. The results of this study may inform global healthimpact assessments at the mid-range of the exposure-response curve. In addition, the study contributes to thegrowing literature on the health effects of air pollution inAsia, particularly given the lack of studies in SoutheastAsia (HEI International Scientific Oversight Committee2010).

Acknowledgements This Technical Assistance was supported withfunds from the Health Effects Institute’s Public Health and AirPollution in Asia (PAPA) Program, the Poverty Reduction Coopera-tion Fund of the Asian Development Bank, (Technical Assistance4714-VIE) and in-kind support from the Government of Viet Nam.The Working Group is grateful to the Clean Air Initiative for AsianCities (CAI-Asia), for initiating communications between HEI, ADB,and the Government of Viet Nam, the local steering committee for theproject, and the (PAPA) program’s International Scientific OversightCommittee for providing technical guidance and suggestions through-out the process, especially Drs. Michael Brauer, Ross Anderson, KirkSmith, and Frank Speizer. We are grateful for useful commentsprovided by HEI’s Review Committee and external quality assuranceconsultant David Bush, and for administrative assistance provided byTiffany North and Morgan Younkin.

Air Qual Atmos Health (2013) 6:167–179 177

References

Barnett AG, Williams GM, Schwartz J, Neller AH, Best TL,Petroeschevsky AL, Simpson RW (2005) Air pollution and childrespiratory health: a case-crossover study in Australia and NewZealand. Am J Respir Crit Care Med 171(11):1272–1278

Braga AL, Saldiva PH, Pereira LA, Menezes JJ, Conceicao GM, LinCA, Zanobetti A, Schwartz J, Dockery DW (2001) Health effectsof air pollution exposure on children and adolescents in SaoPaulo, Brazil. Pediatr Pulmonol 31(2):106–113

Cohen A, Anderson HR, Ostro B, Pandey KD, Krzyzanowski M,Kuenzli N, Gutschmidt K, Pope CA, Romieu I, Samet JM, SmithKR (2004) Mortality impacts of urban air pollution. In: EzzatiMLA, Rodgers A, Murray CJL (eds) Comparative quantificationof health risks: global and regional burden of disease due toselected major risk factors vol 2. World Health Organization,Geneva

D’Ippoliti D, Forastiere F, Ancona C, Agabiti N, Fusco D, MichelozziP, Perucci CA (2003) Air pollution and myocardial infarction inRome—a case-crossover analysis. Epidemiology 14(5):528–535

Ghio AJ (2004) Biological effects of Utah Valley ambient air particlesin humans: a review. J Aerosol Med Depos Clear Effects Lung 17(2):157–164

Ghio AJ, Cohen MD (2005) Disruption of iron homeostasis as amechanism of biologic effect by ambient air pollution particles.Inhal Toxicol 17(13):709–716

Gouveia N, Fletcher T (2000) Respiratory diseases in children andoutdoor air pollution in Sao Paulo, Brazil: a time series analysis.Occup Environ Med 57(7):477–483

Han X, Naeher LP (2006) A review of traffic-related air pollutionexposure assessment studies in the developing world. Environ Int32(1):106–120

Health Effects Institute (2006) PAPA-SAN Database. www.healtheffects.org

HEI International Scientific Oversight Committee (2004) Healtheffects of outdoor air pollution in developing countries of Asia:a literature review. Health Effects Institute, Boston

HEI International Scientific Oversight Committee (2010) Outdoor airpollution and health in the developing countries of Asia: acomprehensive review. Special report 18. Health Effects Institute,Boston

HEI Panel on the Health Effects of Traffic-Related Air Pollution(2009) Traffic-related air pollution: a critical review of theliterature on emissions, exposure, and health effects. HEI specialreport 17. Health Effects Institute, Boston

Hernandez-Cadena L, Barraza-Villarreal A, Ramirez-Aguilar M,Moreno-Macias H, Miller P, Carbajal-Arroyo LA, Romieu I(2007) Infant morbidity caused by respiratory diseases and itsrelation with the air pollution in Juarez City, Chihuahua, Mexico.Salud Publica Mex 49(1):27–36

Janes H, Sheppard L, Lumley T (2005a) Case-crossover analyses ofair pollution exposure data: referent selection strategies and theirimplications for bias. Epidemiology 16(6):717–726

Janes H, Sheppard L, Lumley T (2005b) Overlap bias in the case-crossover design, with application to air pollution exposures. StatMed 24(2):285–300

Karr C, Lumley T, Shepherd K, Davis R, Larson T, Ritz B, Kaufman J(2006) A case-crossover study of wintertime ambient air pollutionand infant bronchiolitis. Environ Heal Perspect 114(2):277–281

Karr CJ, Rudra CB, Miller KA, Gould TR, Larson T, SathyanarayanaS, Koenig JQ (2009) Infant exposure to fine particulate matterand traffic and risk of hospitalization for RSV bronchiolitis in aregion with lower ambient air pollution. Environ Res 109(3):321–327

Kelly FJ (2003) Oxidative stress: its role in air pollution and adversehealth effects. Occup Environ Med 60(8):612–616

Kim JJ, Huen K, Adams S, Smorodinsky S, Hoats A, Malig B, LipsettM, Ostro B (2008) Residential traffic and children’s respiratoryhealth. Environ Health Perspect 116(9):1274–1279

Levy D, Lumley T, Sheppard L, Kaufman J, Checkoway H (2001)Referent selection in case-crossover analyses of acute healtheffects of air pollution. Epidemiology 12(2):186–192

Lin DY, Psaty BM, Kronmal RA (1998) Assessing the sensitivity ofregression results to unmeasured confounders in observationalstudies. Biometrics 54(3):948–963

Lumley T, Levy D (2000) Bias in the case-crossover design: implicationsfor studies of air pollution. Environmetrics 11(6):689–704

Mehta S, Shin H, Burnett R, North T, Cohen AJ (2011) Ambientparticulate air pollution and acute lower respiratory infections: asystematic review and implications for estimating the globalburden of disease. Air Quality Atmos Health. doi:10.1007/s11869-011-0146-3

Mittleman MA (2005) Optimal referent selection strategies in case-crossover studies—a settled issue. Epidemiology 16(6):715–716

Romieu I, Samet JM, Smith KR, Bruce N (2002) Outdoor airpollution and acute respiratory infections among children indeveloping countries. J Occup Environ Med 44(7):640–649

Rothman KJ, Greenland S (1998) Modern epidemiology. Lippencott-Raven, Philadelphia

Saldiva PH, Lichtenfels AJ, Paiva PS, Barone IA, Martins MA,Massad E, Pereira JC, Xavier VP, Singer JM, Bohm GM (1994)Association between air pollution and mortality due to respira-tory diseases in children in Sao Paulo, Brazil: a preliminaryreport. Environ Res 65(2):218–225

Segala C, Poizeau D, Mesbah M, Willems S, Maidenberg M (2008)Winter air pollution and infant bronchiolitis in Paris. Environ Res106(1):96–100

Smith KR, Samet JM, Romieu I, Bruce N (2000) Indoor air pollutionin developing countries and acute lower respiratory infections inchildren. Thorax 55(6):518–532

Smith K, Mehta S, Maeusezahl-Feuz M (2004) Indoor air pollutionfrom household use of solid fuels. In: LA Ezzati M, Rogers A,Murray CJL (eds) Global and regional burden of diseaseattributable to selected major risk factors, vol 2. World HealthOrganization, Geneva

Thomas P, Zelikoff J (1999) Air pollutants: modulators of pulmonaryhost resistance against infection. In: Holgate ST, Samet JM,Koren HS, Maynard RL (eds) Air pollution and health.Academic, San Diego

U.S. Department of Health and Human Services (2006) The healthconsequences of involuntary exposure to tobacco smoke: a reportof the Surgeon General. U.S. Department of Health and HumanServices, Centers for Disease Control and Prevention, Coordi-nating Center for Health Promotion, National Center for ChronicDisease Prevention and Health Promotion, Office on Smokingand Health, Atlanta, GA

Viet Nam Register (2002) Integrated action plan to reduce vehicleemissions in VietNam. ADB. http://www.adb.org/Vehicle-Emissions/actionviet.asp

WHO-CAH (1997) The management of childhood illness in developingcountries: rationale for an integrated strategy. WHO, Department ofChild and Adolescent Health and Development, Geneva

Williams B, Gouws E, Boschi-Pinto C, Bryce J, Dye C (2002)Estimates of world-wide distribution of child deaths from acuterespiratory infections. Lancet 2:25–32

World Health Organization (2004) The global burden of disease: 2004update. WHO, Department of Health Statistics and Informatics inthe Information, Evidence and Research Cluster Geneva, Swit-zerland

178 Air Qual Atmos Health (2013) 6:167–179

Zelikoff JT, Schermerhorn KR, Fang KJ, Cohen MD, Schlesinger RB(2002) A role for associated transition metals in the immunotox-icity of inhaled ambient particulate matter. Environ Heal Perspect110:871–875

Zelikoff JT, Chen LC, Cohen MD, Fang KJ, Gordon T, Li Y,Nadziejko C, Schlesinger RB (2003) Effects of inhaled ambientparticulate matter on pulmonary antimicrobial immune defense.Inhal Toxicol 15(2):131–150

Air Qual Atmos Health (2013) 6:167–179 179