Air pollution and childhood mortality.
The possible impact of air pollution on children’s health was first connected to early child mortality. One of the earliest reports was based on an ecologic study of counties of England and Wales in 1958–1964, with air pollution estimated from indices of domestic and industrial pollution (Collins et al. 1971
). The study found significant correlations between air pollution and infant mortality, and infant respiratory mortality in particular. The Nashville Air Pollution Study, conducted in the 1950s, indicated that dust fall, a measure of air pollution estimated for each census tract, was related to neonatal deaths with signs of prematurity, but the results were inconclusive (Sprague and Hagstrom 1969
). Another early signal that air pollution may be associated with deaths in infancy came from the extensive analyses of air pollution and mortality in 117 U.S. metropolitan areas in the 1960s (Lave and Seskin 1977
). Particulates and, to a lesser degree, sulfate concentrations were positively associated with infant mortality; a 10% increase in pollution was associated with a 1% increase in infant mortality.
Almost two decades passed before a new generation of studies addressed this question in more detail. These newer studies, summarized in , confirmed the early results. A small ecologic study in the Rio de Janeiro, Brazil, metropolitan area reported a positive association between annual levels of particulates and infant mortality from pneumonia (Penna and Duchiade 1991
Air pollution and child mortality.
Bobak and Leon (1992)
studied infant mortality in an ecologic study in the Czech Republic. The study found an association between sulfur dioxide and total suspended particles (TSP), and infant mortality, after controlling for a number of potential confounding variables (at the ecologic level). The effects were specific to respiratory mortality in the postneonatal period. These results were later confirmed in a nationwide case–control study based on the Czech national death and birth registers; this design allowed controlling for social and biologic covariates at the individual level (Bobak and Leon 1999a
). The study found a strong effect of SO2
and TSP on postneonatal mortality from respiratory causes: the relative risks (RRs) per 50 μg/m3
increase in pollutant concentration were 1.95 [95% confidence interval (CI), 1.09–3.50] for SO2
and 1.74 (95% CI, 1.01–2.98) for TSP.
Woodruff et al. (1997)
analyzed the association between early postneonatal mortality and PM10
(particulate matter < 10 μm) levels in about 4 million babies born from 1989 through 1991 in the United States. Infants were categorized as having high, medium, or low exposures based on tertiles of PM10
. After adjustment for other covariates, for total postneonatal mortality for the high-exposure (> 40 μg/m3
) versus low-exposure (< 28 μg/m3
) groups was 1.10 (95% CI, 1.04–1.16). In infants of normal birth weight, high PM10
exposure was associated with respiratory deaths (RR, 1.40; 95% CI, 1.05–1.85) and sudden infant death syndrome (RR, 1.26; 95% CI, 1.14–1.39). The results were similar in term and LBW infants.
Pereira et al. (1998)
investigated the associations between daily counts of intrauterine mortality in the city of São Paulo, Brazil, in 1991–1992 and daily measurements of several pollutants: nitrogen dioxide, SO2
, carbon monoxide, ozone, and PM10
. The association was strongest for NO2
< 0.01). A significant association was also observed with exposure combining the pollutants NO2
, and CO together (p
Loomis et al. (1999)
conducted a time-series study of infant mortality in the southwestern part of Mexico City in 1993–1995. Exposure included NO2
, and particulate matter < 2.5μm (PM2.5
). A 10 μg/m3
increase in the mean level of fine particles during the preceding 3 days was associated with a 6.9% (95% CI, 2.5–11.3%) excess increase in infant death.
Dolk et al. (2000)
examined perinatal and infant outcomes in populations residing near 22 coke works in Great Britain. Data on specific pollutants were not provided; the exposure was based on the proximity to pollution sources. The ratios of observed to expected cases for residence in proximity of the coke works were 0.94 (95% CI, 0.78–1.12) for stillbirth, 0.95 (95% CI, 0.83–1.09) for infant mortality, 0.86 (95% CI, 0.72–1.03) for neonatal mortality, 1.10 (95% CI, 0.90–1.33) for postneonatal mortality, 0.79 (95% CI, 0.30–1.46) for respiratory postneonatal mortality, and 1.07 (95% CI, 0.77–1.43) for sudden infant death syndrome in the postneonatal period. This study, however, had a limited statistical power owing to the relatively small size of the study.
A time-series analysis of daily deaths in Seoul, South Korea, found a relatively specific association between PM10
and total and respiratory mortality in the postneonatal period; the RRs per 10 μg/m3
were 1.03 (1.02–1.04) and 1.18 (1.14–1.21), respectively (Ha et al. 2003
The consistency of these studies, conducted in a range of different populations and using both spatial and time-series study designs, is remarkable. The three largest studies produced very similar estimates of RR (Bobak and Leon 1992
; Woodruff et al. 1997
). Perhaps the only alternative explanation that may affect the interpretation of these studies is confounding by maternal smoking. It is likely that maternal smoking is associated with children’s risk of respiratory death, and none of the studies was able to control for maternal smoking on the individual level. However, at least three observations argue against this possibility. First, all recent studies controlled for socioeconomic factors and other potential confounders. Because smoking in industrialized countries is strongly socially patterned, adjustment for socioeconomic factors should at least partly adjust for smoking. This would be reflected by adjusted estimates being substantially smaller than unadjusted ones. However, in most instances the differences between the crude and adjusted effect estimates were minimal. This does not suggest a presence of residual confounding.
Second, the results of spatial and time-series studies were similar. It is highly unlikely that the social composition or maternal smoking in the studied populations would change substantially over the relatively short periods covered by the time-series studies. In our view, the time-series design practically precludes a presence of confounding by socioeconomic factors or maternal smoking.
Finally, the studies were conducted in very different populations, ranging from China to the United States and from Brazil to the Czech Republic; it is unlikely that the distribution of socioeconomic disadvantage or maternal smoking with respect to air pollution would be similar enough in these different countries to produce the same pattern of results. We therefore conclude that the evidence is sufficient to infer causal relationship between particulate air pollution and respiratory deaths in the postneonatal period.
Air pollution and birth weight.
The potential effects of air pollutants on birth weight were first examined in a small case–control study by Alderman et al. (1987)
; the study did not find any relationship between neighborhood ambient levels of CO during the third trimester of pregnancy and LBW. Over the last decade, this question has been investigated in a number of studies (summarized in ).
Air pollution and birth weight.
Wang et al. (1997)
examined the effects of SO2
and TSP on birth weight in a time-series study in four relatively highly polluted residential areas of Beijing, China. A spectrum of potential confounding factors was adjusted for in multivariate analysis. A graded dose–effect relationship was found between maternal exposure to SO2
and TSP during the third trimester and infant birth weight. The mean birth weight was reduced by 7.3 and 6.9 g for each 100-μg/m3
increase in SO2
and TSP, respectively. The RRs of LBW associated with a 100-μg/m3
increase in SO2
and TSP were 1.11 (95% CI, 1.06–1.16) and 1.10 (95% CI, 1.05–1.14), respectively. The authors speculated that SO2
and particles, or some complex mixtures associated with these pollutants, during late gestation contributed to the LBW risk in the studied population.
Bobak and Leon (1999b)
conducted an ecologic study of LBW and levels of nitrous oxides (NOx
, and TSP in 45 districts of the Czech Republic in 1986–1988. After controlling for socioeconomic factors, the RRs of LBW associated with an increase of 50 μg/m3
in the annual mean concentrations were 1.04 (95% CI, 0.96–1.12) for TSP, 1.10 (95% CI, 1.02–1.17) for SO2
, and 1.07 (95% CI, 0.98–1.16) for NOx
. When all pollutants were included in one model, only SO2
remained related to LBW [odds ratio (OR), 1.10; 95% CI, 1.01–1.20].
In a subsequent study, Bobak (2000)
analyzed individual-level data on all single live births in the Czech Republic that occurred in 1991 in the 67 districts where at least one pollutant (NOx
, or TSP) was monitored. The risk of LBW was analyzed separately for each trimester of pregnancy. The association between LBW and pollution was strongest for pollutant levels during the first trimester of pregnancy. The RRs of LBW per 50 μg/m3
increase in the mean concentration of SO2
and TSP during the first trimester were 1.20 (95% CI, 1.11–1.30) and 1.15 (95% CI, 1.07–1.24), respectively.
In a population-based study in Southern California, Ritz and Yu (1999) examined the influence of pollution levels during the third trimester on LBW risk in a cohort of 126,000 term births. The exposure to O3, NO2, and PM10 in the last trimester was estimated from continuous monitoring data. After adjustment for potential confounders, the risk of LBW was associated with maternal exposure to > 5.5 ppm CO during the third trimester (RR, 1.22; 95% CI, 1.03–1.44). The association between LBW risk and pollution exposure during earlier gestational stages was not significant.
In a population-based case–control study in Georgia (USA), Rogers et al. (2000)
analyzed the combined effect on very low birth weight (VLBW) (< 1,500 g) of SO2
and TSP levels, using annual exposure estimates. The risk of VLBW was increased in babies of mothers who were exposed to concentrations of the combined pollutants above the 95th percentile of the exposure distribution (56.8 μg/m3
); the RR was 2.88 (95% CI, 1.16–7.13).
Maisonet et al. (2001)
examined the association between term LBW and ambient levels of SO2
, and CO in six large cities in the northeastern United States. Their results suggested that the effects of ambient CO and SO2
on the risk of term LBW may differ by ethnic group. In Caucasians (n
~ 36,000), the risk of LBW associated with a 10-ppm increase in SO2
was 1.18 (95% CI, 1.12–1.23) in the first trimester, 1.18 (95% CI, 1.02–1.35) in the second, and 1.20 (95% CI, 1.06–1.36) in the third. By contrast, in African Americans (n
~ 47,000) LBW was associated with CO; a 1-ppm increase in CO concentration was associated with an RR of 1.43 (95% CI, 1.18–1.74) in the first trimester and 1.75 (95% CI, 1.50–2.04) in the third trimester. No effects were seen in Hispanics (n
~ 13,000), although this may have been due to a lower statistical power in this group.
Lin et al. (2001b)
compared the rates of adverse pregnancy outcomes in an area polluted by the petrochemical industry and in a control area in Taiwan. The exposed and control areas differed substantially in the levels of air pollution; for example, the differences in the mean concentrations of PM10
was 26.7 μg/m3
. The RR of term LBW, when the petrochemical municipality was compared with the control municipality, was 1.77 (95% CI, 1.00–3.12).
Ha et al. (2001)
examined full-term births from 1996 through 1997 in Seoul, South Korea, to determine the association between LBW and exposure to CO, SO2
, TSP, and O3
in the first and third trimesters. They found that ambient CO, SO2
, and TSP concentrations during the first trimester of pregnancy were associated with LBW; the RRs were 1.08 (95% CI, 1.04–1.12) for CO, 1.06 (95% CI, 1.02–1.10) for SO2
, 1.07 (95% CI, 1.03–1.11) for NO2
, and 1.04 (95% CI, 1.00–1.08) for TSP.
Vassilev et al. (2001b)
used U.S. Environmental Protection Agency (EPA) Cumulative Exposure Project data to investigate the association between outdoor airborne polycyclic organic matter (POM) and adverse reproductive outcomes in New Jersey for newborns born in 1991–1992. The RR of LBW in term babies, comparing the highest and the lowest exposure groups, was 1.31 (95% CI, 1.21–1.43).
Bobak et al. (2001)
tested the hypothesis that air pollution is related to LBW on data from a British 1946 cohort. They found a strong association between birth weight and air pollution index based on coal consumption. After controlling for a number of potential confounding variables, babies born in the most polluted areas (annual mean concentration of smoke > 281 μg/m3
) were on average 82 g (95% CI, 24–140) lighter than those born in the areas with the cleanest air (mean smoke concentration < 67 μg/m3
Chen et al. (2002)
examined the association between PM10
, CO, and O3
and birth weight in northern Nevada (USA) from 1991 through 1999. The results suggested that a 10-μg/m3
increase in the mean PM10
concentrations during the third trimester of pregnancy was associated a reduction in birth weight of 11 g (95% CI, 2.3–19.8).
Wilhelm and Ritz (2003)
studied the effect on LBW of residential proximity to heavy traffic in Los Angeles County, California (USA) in 1994–1996. The risk of term LBW increased by 19% for each 1 ppm increase in the mean annual concentration of background CO. In addition, an elevated risk was observed for women whose third trimester fell during the fall/winter months (RR, 1.39; 95% CI, 1.16–1.67); this is probably due to the more stagnant air conditions during the winter period. Overall, the study reported an approximately 10–20% increase in the risk of term LBW in infants born to women exposed to high levels of traffic-related air pollution.
A time-series study in Vancouver, Canada, found that LBW was associated with SO2
in the first month of pregnancy (OR per 5 ppb increase, 1.11; 95% CI, 1.01–1.22); NO2
, CO, and O3
were not independently associated with LBW (Liu et al. 2003
). Particles were not measured.
A time-series study in Sao Paulo, Brazil, found that birth weight was inversely related to CO in the first trimester; after controlling for potential confounders, a 1-ppm increase in the mean CO concentration in the first trimester was associated with a 23-g (95% CI, 5–41 g) reduction in birth weight (Gouveia et al. 2004
The results of studies of outdoor exposures are complemented by studies of indoor and personal exposures (not included in ). Boy et al. (2002)
compared the association between birth weight and the type of fuel (open fires with wood smoke, chimney stove, and electricity/gas) used in kitchens by mothers in rural Guatemala during pregnancy. The use of open fire produced average levels of 24-hr PM10
of about 1,000 μg/m3
. Babies of mothers using wood fuel and open fires were on average 63 g (95% CI, 0.4–126 g) lighter than those of women using electricity/gas. Perera et al. (2003)
evaluated the effects of prenatal exposure to airborne carcinogenic polycyclic aromatic hydrocarbons (PAHs) monitored during pregnancy by personal air sampling in a sample of 263 nonsmoking African-American and Dominican women in New York. The mean total PAH exposure was 3.7 ng/m3
(range, 0.4–36.5 ng/m3
). Among African Americans, high prenatal exposure to PAHs was associated with lower birth weight (p
= 0.003) and smaller head circumference (p
= 0.01). No such effects were observed among Dominican women.
Several methodologic issues should be considered in the interpretation of these studies. First, could chance (random error) play a role here? In several of the studies reviewed above, there is a potential problem of multiple comparisons. The more comparisons that are made, the higher the probability that some of them will be “statistically significant.” In some instances, a more stringent use of statistical testing would be useful. Especially in studies where exposures to different pollutants at different pregnancy periods were analyzed, some of the associations could be chance findings. In addition, exposures in different pregnancy periods and concentrations of different pollutants are mutually correlated, and efforts to separate their effects are not reliable.
Second, as with infant mortality, confounding by socioeconomic factors and maternal smoking could affect the results. Overall, however, this seems unlikely for the same reasons as those listed above in “Air pollution and childhood mortality.” In addition, some of the studies were able to control for social conditions and maternal smoking at the individual level, and the results were essentially identical.
In terms of the magnitude of the effect, the results were consistent in suggesting that the effects are relatively small. For comparison, it has been estimated that active smoking in pregnancy leads to a reduction in birth weight by approximately 150–200 g (Adriaanse et al. 1996
), and exposure to environmental tobacco smoke in pregnancy results in birth weight reduction by approximately 20–30 g (Windham et al. 1999
). There were also substantial inconsistencies in the results with respect to the importance of individual pollutants and the timing of critical exposure. The extent of the inconsistencies was such that the studies were not “combinable” into a formal meta-analysis to produce pooled effect estimates, although it is possible that the mix of pollutants differs between different settings and that this underlies the discrepancies in results.
The evidence suggests causality of the effect of air pollution on birth weight. However, given the potential problem with multiple comparisons and the heterogeneity of results, further studies are needed to confirm that the effect is indeed causal, to clarify the most vulnerable periods of pregnancy and the role of individual pollutants, and to examine whether the impaired reproductive outcomes have any long-term consequences on child health.
Air pollution and premature births.
Perhaps the first study that suggested a possible association between air pollution and preterm births was the Nashville Air Pollution Study; the results suggested that dust fall (a measure of particulate pollution) was associated with neonatal deaths among premature births (Sprague and Hagstrom 1969
). However, the study did not address the question of preterm births specifically, and there were concerns about confounding by socioeconomic variables. It was only in the 1990s when this issue was investigated in more detail ().
Air pollution and premature births.
The first “modern” investigation of the possible influence of air pollution on premature birth was a time-series study in Beijing, China, conducted by Xu et al. (1995)
. The study found an inverse relationship between gestational age and the concentration of SO2
and TSP; the RRs of premature birth associated with a 100-μg/m3
increase in the mean SO2
and TSP concentrations during pregnancy, after controlling for potential confounders, were 1.21 (95% CI, 1.01–1.45) and 1.10 (95% CI, 1.01–1.20), respectively. Trimester-specific effects were not studied.
examined the relation between premature birth and ambient NOx
, and TSP during each trimester. The association was strongest for SO2
, weaker for TSP, and only marginal for NOx
. For exposures during the first trimester, the RRs associated with a 50-ng/m3
increase in pollutant concentrations were 1.27 (95% CI, 1.16–1.39) and 1.18 (95% CI, 1.05–1.31) for SO2
and TSP, respectively. The effects of pollutants on premature births in the later two trimesters were weak.
The possible impact of CO, NO2, O3, and PM10 on premature birth was studied by Ritz et al. (2000) in Southern California. After adjustment for a number of biologic, social, and ethnic covariates, premature births were associated with CO and PM10 in the first gestational month and during late pregnancy. The RR associated with PM10 during the first gestational month was 1.16 (95% CI, 1.06–1.26); exposure in the last 6 weeks of gestation was associated with an RR of 1.20 (95% CI, 1.09–1.33). The association of premature birth with CO levels was not consistent throughout the study area.
In a study in a petrochemically polluted area in Taiwan, Lin et al. (2001a)
found an RR of preterm birth in the polluted area, compared with the clean area, of 1.41 (95% CI, 1.08–1.82), after controlling for potential confounders.
The Vancouver time-series study found that the risk of preterm birth was associated with SO2
and CO during the last month of pregnancy; the ORs were 1.09 (1.01–1.19 per 5-ppb increase) and 1.08 (1.01–1.15 per 1-ppm increase), respectively (Liu et al. 2003
The interpretation of the studies of preterm birth is complicated by similar issues as in the case of birth weight: the issue of multiple comparisons, and the inconsistency of the results in terms of the role of individual pollutants and the timing of exposure. In addition, there have been fewer studies of premature birth than of birth weight. We therefore conclude that the evidence, as yet, is insufficient to infer causality, but further studies are justified.
Air pollution and IUGR.
IUGR is defined as birth weight below the 10th percentile of birth weight for gestational age and sex. IUGR is an interesting end point that may predict functional changes in adulthood, such as hypertension and coronary heart disease. The studies of the relationship between IUGR and air pollution are summarized in .
Dejmek et al. (1999)
examined the impact of PM10
on IUGR in a highly polluted area of northern Bohemia (Teplice District). The mean concentration of pollutants in each month of gestation for each mother were estimated from continuous monitoring data. A significantly increased risk of giving birth to a child with IUGR was established for mothers who were exposed to PM10
levels > 40 μg/m3
> 27 μg/m3
during the first month of gestation. The adjusted odds ratio (AOR) associated with a 10-μg/m3
increase in PM10
was 1.25 (95% CI, 1.08–1.56); a similar, although weaker, association was seen for PM2.5
. There was no association between IUGR and particulate levels in later gestational months or with SO2
, or O3
(Dejmek et al. 1996
The question of IUGR was addressed again in a reanalysis of an extended data set (Dejmek et al. 2000
). Compared with exposure to the mean PM10
of < 40 μg/m3
during the first month of gestation, the AOR was 1.44 (95% CI, 1.03–2.02) for the medium-exposure group (PM10
40 to < 50 μg/m3
) and 2.14 (95% CI, 1.42–3.23) for PM10
of ≥ 50 μg/m3
. Using a continuous exposure, the AOR of IUGR was 1.19 (CI, 1.06–1.33) per 10-μg/m3
increase of PM10
in the first gestational month.
In further analyses of this cohort, Dejmek et al. (2000)
investigated the association between carcinogenic PAHs and IUGR in two Czech districts: Teplice and Prachatice. In the Teplice data, there was a highly significant increase of IUGR with exposures to carcinogenic PAHs (benz[a
]pyrene, chrysene, dibenz[a
]anthracene, and indeno[1,2,3-c
]pyrene) above 15 ng/m3
. Again, the effect was specific for the first gestational month. The AORs were 1.59 (95% CI, 1.06–2.39) for medium levels of carcinogenic PAHs and 2.15 (95% CI, 1.27–3.63) for high exposure levels. Using a continuous measure of exposure, a 10 ng/m3
increase in carcinogenic PAH level was associated with an AOR of 1.22 (95% CI, 1.07–1.39). Although there was no effect of PM10
on IUGR found in Prachatice, the association between carcinogenic PAHs and IUGR was close to that found in Teplice. Again, the only consistent association between carcinogenic PAHs and IUGR was observed in the first gestational month: compared with the lowest category of exposure to carcinogenic PAHs, the AOR of IUGR was 1.63 (95% CI, 0.87–3.06) in the medium category and 2.39 (95% CI, 1.01–5.65) in the highest category.
The analysis of the Czech national birth register linked with air pollution data did not reveal any significant association between IUGR and ambient levels of NOx
, and TSP (Bobak 2000
). The discrepancy between the Czech studies is probably related to exposure measurement. PAHs appear to be the critical exposure for IUGR, but PAHs were not measured by the national monitoring system used for exposure estimation by Bobak (2000)
Vassilev et al. (2001a)
examined the association of POM in outdoor air with “small for gestational age” (SGA) births (the definition of SGA is identical to that of IUGR). Information from birth certificates in New Jersey (USA) from 1991 through 1992 was combined with data on air toxicity derived from the U.S. EPA Cumulative Exposure Project, using the predicted POM concentrations from annual exposure estimates. The AOR for term SGA in the highest exposure tertile (0.61–2.83 μg/m3
, which includes about 89% of the state’s births) was 1.22 (95% CI, 1.16–1.27), suggesting that residential exposure to airborne POM is associated with an increased prevalence of IUGR.
In the Vancouver study, using the time-series approach, SO2
, and CO in the first month of pregnancy were associated with IUGR; the ORs were 1.07 (95% CI, 1.01–1.13) per 5-ppb increase, 1.05 (95% CI, 1.01–1.10) per 10-ppb increase, and 1.06 (95% CI, 1.01–1.10) per 1-ppm increase, respectively (Liu et al. 2003
). Data on exposures to particles or PAHs were not available in that study.
As with studies of birth weight and preterm births, the reviewed studies of IUGR produced inconsistent results, and the interpretation is complicated by multiple comparisons (Bobak 2000
; Liu et al. 2003
) and mutual correlations of exposures. The results by Dejmek et al. (1999
and Liu et al. (2003)
suggest that the first month was the most sensitive period for the effect of air pollutants, but further studies should clarify this issue. Data by Dejmek et al. (2000)
and Vassilev et al. (2001a)
imply a critical role of PAHs. It is possible that carcinogenic PAHs are responsible for the biologic activity of complex mixtures adsorbed to respirable air particles that can result in IUGR. With the increase in traffic, the significance of PAHs in Europe is growing, but their monitoring remains scarce. At present, the evidence is insufficient to infer causality, but further studies are required.