METHODS

Data on live births were obtained from the Birth Information System (SINASC) of the Ministry of Health in Brazil. According to this system every live birth is expected to result in a certificate, which is later requested for legal registration of the child. These certificates are captured on local registers and sent to the State Office for Statistics where they are compiled.

For each individual birth, information from these certificates is available on maternal age, maternal education, place of residence, parity, number of antenatal care visits, type of pregnancy (single, double, triple or more), length of gestation in weeks, date of birth, site of birth (home, hospital or other health service), method of delivery (spontaneous, surgical or forceps), Apgar scores, birth weight, and sex.

According to recent evaluations this information system is considered to be adequate and capable of providing good quality information for the main variables, especially for the city of São Paulo.²¹ All births to mothers resident in the city of São Paulo during 1997 were obtained from this register.

Data on air pollution were available from CETESB, the environmental agency in São Paulo. These data comprised daily means (24 hour averages) of SO₂ and respirable particles (PM₁₀), the maximum hourly mean calculated during the 24 hour period for NO₂ and ozone (O₃), and the maximum eight hour moving average for CO. These data were collected from a minimum of four monitoring sites (CO) and a maximum of 12 sites (PM₁₀ and SO₂). For each pollutant, as inter-site correlation was high, daily citywide levels were calculated by averaging all available data across all monitoring stations. The attainments of the network of monitoring stations and the data handling capabilities of CETESB have been considered to be of high quality.²²

The analysis of the relation between birth weight and air pollution was carried out only for live singleton full term pregnancies (⩾37 weeks of gestation). Birth weight was considered in separate analyses as a (1) continuous and as a (2) binary variable taking the values of 1 for birth weight of less than 2500 g and 0 otherwise.

Maternal exposure to air pollution at each trimester of the pregnancy was estimated taking into account the date of birth and the length of each gestation. Mean exposures of about three months to each air pollutant were then assigned to each birth. Logistic regression was used to identify maternal and pregnancy factors associated with low birth weight.

By way of regression using GAM models²³ to allow for the non-linear effect of maternal age on birth weight we assessed the change in birth weight associated with the mean exposure to air pollution during each trimester of pregnancy adjusting for the factors related to low birth weight. We used logistic regression to estimate the odds of having a low birthweight baby according to maternal exposure while adjusting for the same factors. Single and multiple pollutant models were examined. All analyses were carried out in S-Plus 2000 (Mathsoft, Seattle, WA). Care was taken in relation to the model convergence criteria and in computing exact standard errors of the pollutant estimates.²⁴

RESULTS

There were 214 973 live births to mothers resident in the city of São Paulo during 1997. Of these, 196 978 (91.6%) were singleton pregnancies with complete information for gestational age. Pre-term live births accounted for 6.2% of these pregnancies and were excluded from the analysis. In addition, we decided to further exclude very low (<1000 g) and very high (>5500 g) birthweight babies as we assumed that those babies were very likely to be the product of high risk pregnancies. Thus, any possible effect of air pollution on those pregnancies would be outweighed by other high risk conditions. This resulted in a population of 179 460 live births for the final analysis.

Figure 1 shows the daily mean birth weight during 1997 in São Paulo. Although our data were restricted to just one year, it can be observed that birth weight exhibited a seasonal pattern. This was confirmed by the descriptive analysis of birth weight in table 1, which also depicts the seasonal pattern of air pollution during the period of this study.

The mean birth weight, the proportion of low birthweight babies and adjusted odds ratios (OR) for a low birthweight baby according to maternal and pregnancy characteristics are presented in table 2. The mean birth weight among singleton full term pregnancies was 3186 g with an overall proportion of low birth weight (<2500 g) of 5.0%.

Birth weight was shown to be associated with infant gender, maternal age, maternal education, number of antenatal care visits, parity, and the type of delivery. The relation with maternal age is clearly non-linear with greater risk of a low birthweight child for both very young and older mothers.

Although the analysis was restricted to full term pregnancies, birth weight was still significantly associated with length of gestation (37 to 39 weeks compared with 41+weeks). This association was observed first in univariate analyses and persisted in multifactorial models. This variable was thus considered in all multifactorial models.

Multiple regression was used to assess the relation between birth weight and maternal exposure to air pollution allowing for effects of the above variables. To adjust for the potential influence of season on birth weight an indicator variable for month of birth was included. Because of the non-linear relation between maternal age and birth weight, a non-parametric smooth function of maternal age was fitted in the generalised additive model using a spline function.²³

The variables “number of antenatal care visits” and “parity” both had a high percentage of missing observations (around 30%). The inclusion of both variables in the same regression model resulted in a high percentage of observations being left out leading to very imprecise estimates. Thus, we decided to include in the final models only the number of antenatal care visits as this variable explained proportionally more of the variation in birth weight than parity. Exploratory analysis modelling both variables at the same time or including just parity all gave similar results.

Table 3 shows the coefficients obtained in GAM regression models for birth weight expressed as changes in mean birth weight according to a 10 μg/m³ or a 1 ppm change in exposure to the air pollutants in each trimester of pregnancy. All pollutants exhibited an inverse relation with birth weight when exposure occurred in the first trimester. However, only PM₁₀ and CO had statistically significant associations. During the second and third trimester results followed different directions and were mostly not significant. Only SO₂ exhibited a positive and significant relation with birth weight.

Semi-parametric logistic regression models were then explored again using a smooth function to control for the non-linear effect of maternal age. Levels of air pollutants were divided into quartiles and the lowest quartile (<25th) was used as the reference exposure category. Results of this analysis are shown on table 4. Although generally not significant, results of this analysis showed that the pattern of the association between birth weight and PM₁₀ and CO during the first trimester was compatible with the results of the GAM regression—that is, an increased risk of a low birthweight baby for higher exposures to air pollutants. The relation between birth weight and exposure to SO₂ during the second trimester also exhibited an inverse pattern with higher risk of a low birthweight baby.

The last step was to try and establish which pollutant was responsible for the association with low birth weight. Multiple pollutant models were examined only for the pollutants that showed statistically significant associations in single pollutant linear models—that is, PM₁₀, CO and SO₂. The rationale for this analysis is that, considering that the pollutants were highly correlated, if one was acting only as a proxy for the other with no contribution of its own, it would be expected that its parameter would be attenuated in a two pollutant model while that of the other would remain relatively unchanged. Results of this analysis are displayed in table 5.

The inclusion of PM₁₀ and CO (mean exposure of the first trimester) simultaneously in the same model resulted in a reduction in the coefficient for both pollutants. However, the reduction observed for CO (19%) was proportionally smaller than for PM₁₀ (61%). Similar changes happened when CO and SO₂ (mean of the second trimester) were examined together. The coefficients for PM₁₀ and SO₂ had smaller decreases when these two pollutants were examined together. In a three pollutant model the decreases in the size of coefficients were greater but still of lesser magnitude for CO. The high correlation between PM₁₀ and CO (around 0.9) should be noted.

DISCUSSION

This study found some evidence that maternal exposure to air pollution, especially during the first trimester of pregnancy, is associated with reductions in birth weight. Although it remains difficult to separate the effect of each pollutant, it appears that CO and possibly PM₁₀ have an independent and significant negative association with birth weight. On adjusting for infant gender, maternal age, maternal education, antenatal care, type of delivery, month of birth, and seasonal patterns, an increase in mean exposure to CO of 1 ppm during the first trimester was associated with a reduction of about 23 g in birth weight. For a 10 μg/m³ change in PM₁₀ we observed a 13.7 g reduction.

There is still a very limited literature regarding effects of air pollution on birth weight and results of these studies are somewhat controversial. While some studies failed to find an effect of air pollution on birth weight,^25,26 and others have found an effect but with no specification of timing of exposure,^8,27 some have found stronger and more consistent effects when exposure occurred during the third trimester,^9,11,12 while others found effects in the first trimester.^10,13,28,29

The possible biological mechanisms involved in the reduction of birth weight associated with maternal exposure to air pollution vary according to the timing of this exposure. The implantation of the fetus and the formation of the placenta occur during the first trimester while weight gain occurs predominantly during the third trimester.³⁰ Therefore, exposure during both periods presents the possibility of interference with the final birth weight.

Basic concepts of pathophysiology of fetal growth retardation are based on different levels of maternal supply, fetoplacental competition, and fetal adaptations. In recent studies knowledge about placental development and function has been increased.³¹ Abnormalities in placental development may occur during its formation, and cellular and molecular functions may be changed leading to inadequate implantation and growth. Abnormalities in placental transport may also develop later on, because of problems in the uteroplacental circulation, exchange at the intervillous space, and umbilical and fetal circulation. All these factors lead to problems with fetal adaptation mechanisms, most importantly decreased fetal growth rate and fetal activities. In the first trimester genetic mutations are considered to be the most important element in placental abnormalities, and in the second and third trimesters extremely complex vascular alterations are considered the main cause of placental abnormalities and consequent fetal growth retardation. Pollutants are recognised as being able to have an effect on both dimensions.³¹

Our finding of a more consistent effect during the first trimester is coherent with this mechanism. Moreover, our results are supported by findings from Seoul,¹³ from the Czech Republic,^28,29 and from a study in the Treplice district of the Czech Republic.¹⁰ This last study examined exposures during all nine months of pregnancy and revealed that exposure during the first month was the most important for intrauterine growth retardation.

Most air pollutants (CO, SO₂, NO₂, and particles) have already been described as being associated with reductions in birth weight. But among them, CO and particles have been more consistently found in the few studies published.^9–13,28 As pollutants are highly correlated, it is usually difficult to isolate their effects. We tested multiple pollutant models in the present analysis and found that CO and PM₁₀ exhibited significant associations that appeared to be independent, with CO presenting a more consistent effect.

Again this finding is coherent with the biological mechanism behind the observed effects as maternal smoking during pregnancy, the environmental factor that best correlates with low birth weight,³² also produces CO which has toxic and ischaemic effects upon fetal growth.

While our results are in accordance with the limited literature regarding effects of air pollution on pregnancy outcomes, some methodological aspects of this study should be discussed. For example, the birth data available for this study came from a system that is known for the completeness and quality of the information provided. However, information on length of gestation was available only in categories (<22 weeks, 22 to 27, 28 to 31, 32 to 36, 37 to 41, 42 or more weeks). Although this was not a problem for the definition of term and preterm births, the information on length of gestation was used to calculate the mean exposure to air pollution for each trimester of the pregnancy. Therefore, we should expect a certain degree of inaccuracy in the maternal exposure. However, it is probable that such inaccuracy would be non-differential and any bias introduced would only attenuate the effects found.

In addition, information on important known risk factors for a low birthweight baby such as maternal smoking status, maternal associated morbidity, pre-pregnancy weight, and weight gain were not available. Among these risk factors cigarette smoking is considered the most important predictor of low birth weight.³² However, there is some indication that maternal smoking status is associated with maternal education in Brazil³³ and consequently, adjusting for the last variable contributes to partially adjust for cigarette smoking. In addition, there are no reasons to believe that such risk factors would vary according to levels of air pollution, the exposure of interest in our study. Thus, important confounding effects because of lack of information on such variables is not expected.

An important concern in any epidemiological study is the correct assessment of exposure of the subjects being studied. In our study we defined the exposure of each mother according to citywide levels of air pollution during each trimester of pregnancy.

Firstly, we have the averaged period of three months. It could be argued that a smaller time period would be more appropriate or that a measure that takes into account the variability of air pollution levels during each trimester would better represent the exposure. A study in the Czech Republic¹⁰ used monthly averages instead of trimesters. However, we could not find any justification for the use of one or the other measurement of exposure. Most of the studies that assessed pregnancy outcomes in relation to air pollution have used trimester exposures. The trimesters represent a traditional time period for the study of the pregnancy period. In addition, any effect of air pollution on birth weight is likely to result from long term exposure to increased levels rather than short term fluctuations. Thus, the mean over trimesters might be a reasonable indicator. Nevertheless, future studies should attempt to explore and compare different indicators of exposure.

A second problem is the fact that each mother was assigned a citywide level of air pollution. It appears that levels of CO can vary considerably across different areas of a city and thus individual exposure could be underestimated or overestimated when averaging levels of all stations. Ritz and Yu¹¹ in their study in Los Angeles assigned the exposure of mothers according to the closest monitoring station to each mother’s residence. However, we have to consider the possibility that pregnant women move around different areas of a city for reasons such as working and shopping. Thus, their exposure might not be well represented by just one station. In fact, any approach used for exposure assessment will be subject to error, but this error is likely to be non-differential and again its main consequence is the attenuation of the effect estimated.

Notwithstanding, this study has the advantage of examining a considerable number of births occurring in a comparatively small area (there are about 200 000 births annually in the city of São Paulo, which covers an area of 1509 km²). Moreover, besides statistical power and a fairly homogeneous exposure to air pollution, seasonal patterns were taken into account for this analysis. Despite a few caveats, these results are consistent in revealing that exposure to air pollution during pregnancy may interfere with weight gain in the fetus.

Low birth weight is one of the strongest predictors of infant mortality in both developed and developing countries. It is also a more sensitive indicator of environmental effects than late fetal loss or perinatal deaths. Thus, even small effects can be detected in a large population. Therefore, identification of such a widespread risk factor that, to date, has been understudied is very important from the public health perspective. The association between air pollution and birth weight and also other pregnancy outcomes should be confirmed in other locations through well designed studies. Increasing the coherence of this relation will help efforts to minimise human exposure to air pollution in order to protect public health.