Adverse Birth Outcomes Related to NO2 and PM Exposure: European Systematic Review and Meta-Analysis

There is a growing number of international studies on the association between ambient air pollution and adverse pregnancy outcomes, and this systematic review and meta-analysis has been conducted focusing on European countries, to assess the crucial public health issue of this suspected association on this geographical area. A systematic literature search (based on Preferred Reporting Items for Systematic reviews and Meta-Analyses, PRISMA, guidelines) has been performed on all European epidemiological studies published up until 1 April 2020, on the association between maternal exposure during pregnancy to nitrogen dioxide (NO2) or particular matter (PM) and the risk of adverse birth outcomes, including: low birth weight (LBW) and preterm birth (PTB). Fourteen articles were included in the systematic review and nine of them were included in the meta-analysis. Our meta-analysis was conducted for 2 combinations of NO2 exposure related to birth weight and PTB. Our systematic review revealed that risk of LBW increases with the increase of air pollution exposure (including PM10, PM2.5 and NO2) during the whole pregnancy. Our meta-analysis found that birth weight decreases with NO2 increase (pooled beta = −13.63, 95% confidence interval (CI) (−28.03, 0.77)) and the risk of PTB increase for 10 µg/m3 increase in NO2 (pooled odds ratio (OR) = 1.07, 95% CI (0.90, 1.28)). However, the results were not statistically significant. Our finding support the main international results, suggesting that increased air pollution exposure during pregnancy might contribute to adverse birth outcomes, especially LBW. This body of evidence has limitations that impede the formulation of firm conclusions. Further studies, well-focused on European countries, are called to resolve the limitations which could affect the strength of association such as: the exposure assessment, the critical windows of exposure during pregnancy, and the definition of adverse birth outcomes. This analysis of limitations of the current body of research could be used as a baseline for further studies and may serve as basis for reflection for research agenda improvements.


Introduction
Low birth weight (LBW) is defined by the World Health Organization (WHO) as birth weight less than 2500 g (referenced P07.0-P07.1 in the 10th revision of the international classification of diseases-ICD 10) [1]. In addition, preterm birth (PTB) is defined as childbirth occurring at less than 37 completed weeks or 259 days of gestation (referenced P07.2-P07.3 in ICD 10). The WHO estimated promoting cleaner fuel sources and energy technologies, promoting smarter urban planning that aims to reduce urban density and traffic-related pollution, etc. [40]. So far, environmental policies designed to reduce air pollution issue have shown to be effective, with health benefits and helping to reach health policy objectives [41,42]. For instance: Japanese legislation has limited transportation-related emission since 2001. The average NO 2 concentration decreased from 30 to 21 ppb and PM 2.5 concentrations decreased from 38 to 26 mg/m 3 . These reductions respectively led to 1.1% and 0.6% lower prevalence of pediatric asthma [43].
To date, Health Impact Assessments (HIA) are recognized to play a crucial role in evaluating different policy scenarios for reducing air-pollution levels; in assessing new air-quality directives; or in calculating the external monetary costs of air pollution or the benefits of preventive actions [44,45]. More precisely, an HIA in this field provides the number of health events attributable to air pollution in the target population [45] and, thereby, in our case, quantifies the air pollution burden of disease due to adverse birth outcomes as preterm birth and low birth weight complications in Europe [46]. Assessment of environmental burden of disease enable the identification of policy priorities. To implement a HIA, several data sources are needed, including the dose-response function; this function derives from epidemiological studies assessing statistical indicator as relative risk associated with the modelled and observed exposure [47]. In our case, this relative risk may come from Europe based meta-analysis providing pooled estimates. One substantial input of meta-analysis is to offer estimates within a specific vulnerable population as well as a closer match with the geographical context of exposure [48]. More often, the dose-response curve linking air pollution and health impacts is supposed to be linear which means that reductions in air-pollution levels, will have consequences for health effects independently to the starting point on the curve. Therefore, this linear relationship cannot capture the different level of an individual's susceptibility to air pollution [49,50]. It is a reason why preventive action aimed at reducing air-pollution levels in general and not only focusing on air-pollution peaks. Focusing on the peaks of air pollution would only prevent a small number of health events [45].
Recently, there has been a growing number of studies investigating the relationship between adverse birth outcomes, as PTB and LBW, and air pollutant concentration. The possible effect of air pollution exposures on birth outcomes has been reviewed in several systematic reviews and meta-analyses [26,37,48,[51][52][53][54][55][56][57][58]. To the best of our knowledge, no European systematic review was performed to consider more homogeneous level of exposure to air pollution. The European Union and WHO have drafted a legislative framework which establishes health-based standards and objectives for several air pollutants. For instance, the threshold for the particulate matter (PM 10 ) concentrations is 40 µg/m 3 on 1 year, for PM 2.5 25 µg/m 3 on 1 year, for NO 2 40 µg/m 3 on 1 year and for SO2 it is 125 µg/m 3 on 24 h, these regulations differ from one continent to another. In this way, the average concentration of various air pollutants differs from one country to another. For instance, the level of exposure to annual average concentration of NO 2 in the countries of the world, between 2000 and 2015 varied from 97 µg/m 3 (NYC, USA) and 55 µg/m 3 (Beijing, China) into 35 µg/m 3 (Paris, France) and 26.1 µg/m 3 (Valencia, Spain) [59][60][61][62].
In this setting, updating the literature synthesis of European studies may improve our understanding of the relationship between air pollution, and PTB (as well as LBW). Therefore, we conducted a meta-analysis to assess the association between air pollution and the risk of PTB and LBW, separately, in order to suggest future directions for European research and public health policies.
Our work investigated the following epidemiological question: among newborn in European countries, is air pollution exposure of women during pregnancy significantly related to a risk of adverse birth outcome including weight and term of birth in observational studies ? We focused our analysis only the European studies which investigated the relationship between PM and NO2 and birth outcome-LBW and PTB-in order to produce an appropriate dose-response function within a specific European population as well as a closer match with the geographical context of exposure. Therefore, our European meta-analysis could go beyond the main limitation of HIAs performed today to quantify the environmental burden of disease.

Search Strategy
The systematic literature search was conducted with the PubMed platform in order to access to the Academic Search Complete databases and Medline, among articles published up until 1 April 2020. The search strategy followed the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) guidelines [63] and was performed with the following keywords found in article titles and/or abstract: "ambient air pollution" OR "outdoor air pollution" OR "atmospheric air pollution" AND "birth outcomes" OR "pregnancy outcomes" OR "low birth weight" OR "birth weight" OR "low-birth-weight" OR "birthweight" OR "birth-weight" OR "preterm birth" OR "gestational age" OR "LBW" OR "PTB" AND "Europe" OR "European" OR "Austria" OR "Belgium" OR "Bulgaria" OR "Croatia" OR "Cyprus" OR "Czech Republic" OR "Denmark" OR "Estonia" OR "Finland" OR "France" OR "Germany" OR "Greece" OR "Hungary" OR "Ireland" OR "Italy" OR "Latvia" OR "Lithuania" OR "Luxembourg" OR "Malta" OR "Netherlands" OR "Poland" OR "Portugal" OR "Romania" OR "Slovakia" OR "Slovenia" OR "Spain" OR "Sweden" OR "United Kingdom" At the first step, the inclusion criteria were human studies, peer-reviewed papers written in English and articles published after 1998.

Studies Selction Strategy
We restricted our systematic review on geographical location with European study only-for the reason described above-on the pregnant women and pregnancy outcomes, and on ambient air pollution. Papers presenting non-original studies were ultimately excluded.
At the second step, the inclusion criteria were specific pregnancy outcomes definitions including birthweight, low birth weight, preterm birth or small for gestational age (SGA). Secondary criteria were studies investigated specific outdoor air pollutants measured including NO 2, PM 10 , PM 2.5 .
Two authors (VS and WK) independently screened the papers based on information in the title, abstracts and full manuscripts to select those papers considered relevant based on the screening criteria described below At the last step, to perform meta-analysis, among articles included according to the inclusion criteria for the systematic literature review, the inclusion criteria were studies with measure of association between pollutant concentration and birth outcome.
In the last step, bibliographic reference lists of all included studies were searched manually to identify additional studies cited by the previous references.
Finally, meta-analysis was not performed when less than four studies were available for measures of association between a given outcome and a pollutant. Consequently, of the 14 articles included in this systematic literature review, 4 were excluded according to the inclusion criteria for the meta-analysis. Finally, 10 articles were included in the meta-analysis. At the first step, the inclusion criteria were human studies, peer-reviewed papers written in English and articles published after 1998.
We restricted our systematic review on geographical location with European study only-for the reason described above-on the pregnant women and pregnancy outcomes, and on ambient air pollution. Papers presenting non-original studies were ultimately excluded.
At the second step, the inclusion criteria were specific pregnancy outcomes definitions including birthweight, low birth weight, preterm birth or small for gestational age (SGA). Secondary criteria were studies investigated specific outdoor air pollutants measured including NO2, PM10, PM2.5.
Two authors (VS and WK) independently screened the papers based on information in the title, abstracts and full manuscripts to select those papers considered relevant based on the screening criteria described below At the last step, to perform meta-analysis, among articles included according to the inclusion criteria for the systematic literature review, the inclusion criteria were studies with measure of association between pollutant concentration and birth outcome.

Data Extraction
For each study, we extracted and reported in several tables the following information: • Assessments of association including odds ratios (ORs), hazard ratios (HRs), relative risks (RRs) and other metrics measuring the strength of association between outcomes and exposure to different pollutants including NO 2, PM 10 , PM 2.5 were extracted. When several measures of association were available, we reported those one from the fully adjusted models.
The two authors (VS and WK) independently extracted all data from selected studies.

Meta-Analysis
When at least four studies were available, the pooled estimate between pregnancy outcomes and exposure to air pollutant was computed. Studies' risk and beta estimates were expressed as unit corresponding to an increase of 10 µg/m 3 . A fixed or random model based on the Cochran Q-test, the I-square statistic, and the associated p-value, was used to obtain the combined effect. The level of heterogeneity between studies is quantified with the I-square indicator (I 2 ). When the Cochran Q-test do not reveal significant heterogeneity between studies, a fixed model was applied; inversely, a random model was implemented when the Cochran Q-test was significant. Q-test value between 25% and 50% correspond to a low level of heterogeneity, between 50% and 75% a medium level of heterogeneity and >75% corresponds to a high level of heterogeneity. Forest plots were used to visualize the combined risk estimates. Statistical analysis was performed using the STATA 11 software.

Studies Selected for Review
In accordance with criteria summarized in Figure 1, in all 134 published selected, a total of 84 studies were excluded based on titles. At the second step, titles of the 134 were screened by two authors (VS and WK) independently. A total of 84 studies were excluded based on the criteria described above. At the third step, the abstracts of the remaining 50 articles (of the 134 articles initially selected) were thoroughly read independently by two experts (VS and WK, authors of this article); 16 were then excluded following criteria described above.
Full manuscripts of the remaining 30 articles (of the 134 articles initially selected) were thoroughly read and 16 articles were excluded. Finally, a total of 14 articles were included according to the inclusion criteria for the systematic literature review. Finally, bibliographic reference lists of all included studies were searched manually to identify additional studies cited by the previous references. No additional article was found. Selected studies are defined in Table 1.
In order to perform a meta-analysis, studies were excluded where there was with a measure of exposure not expressed as a pollutant concentration (for instance: exposed/not exposed) or without measure of association, or when the outcome or the exposure (NOx in summer season) was not pertinent for the meta-analysis.
At last, meta-analysis was performed when at least four studies were available for measures of association between a given outcome and a pollutant. Consequently, of the 14 articles included in this systematic literature review, 4 were excluded according to the inclusion criteria for the meta-analysis. Finally, 10 articles were included in the meta-analysis.

General Description
There were 30 studies published since 1998, including more than 47,805 low birth weight newborns (and subtypes), 311,432 preterm birth (and subtypes) and 3319 newborns small for gestational age, in order to estimate the association between adverse pregnancy outcomes and exposure to three ambient pollutants, NO 2, PM 10 and PM 2.5 . About 10 were eligible for the meta-analyses with the exclusion of 4 studies [75,81,89,91]. Of these, LBW, VLBW, ELBW, PTB, VPTB, EPTB, SGA, gestational age and birth weight were investigated (Table 1). About 4046 cases of preterm birth were included in the meta-analyses and 12,502 births were used to study the birth weight.
Databases were drawn mainly from birth certificate information and health database from hospital information systems while other form institutes of national health statistics and cohort databases were also used.

Overview of Current Evidence Concerning Possible Effects on Birth Outcomes of Exposure to Air Pollution
In this section, the results of studies are presented in Figures 2-4, structured by window of exposure of different pollutants (NO 2 , PM 10 , PM 2.5 ). Overall, results show the risk of adverse birth outcomes increases for a 10 µg/m 3 increase NO 2 exposure. Therefore, 19 results tend to show an association between the increase of risk of adverse pregnancy outcomes and NO 2 exposure while 10 results which tend to show a decrease of these risks. Our review reveals that for 10 µg/m 3 increase in NO 2 exposure ( Figure 2) newborn have increased risk of:               However, several results were not significant, except studies [64,68,75]. Among studies focusing on critical windows, during each window of exposure the number of results which tend to show an association between PTB or SGA and air pollutant are the same and do not increase or decrease with the trimester of pregnancy, for any windows of exposure only three results tend to show an association [64,68,85].
Whereas the risk of LBW seems to increase as the pregnancy progresses. In this way, our review reveals that two results tend to show an association between the risk of LBW and air pollutant exposure (NO 2 , PM 10 , PM 2.5 , Figures 2-4) during the first trimester of pregnancy (Appendix B), three results tend to show this association during the second trimester of pregnancy (Appendix C) and four results tend to show this association during the third trimester (Appendix D).
In addition, when studies consider the exposure of the entire pregnancy, seven results found an association between air pollutant exposure and the increase of the risk of LBW against only 3 results for PTB and SGA in the same windows (Appendix A).
Among studies focusing on the 1st trimester of exposure the risk of adverse birth outcomes ranges from 0.78 to 1.67 with confidence interval range from 0.53 to 2.18. For the 2nd trimester of exposure results (OR) range from 0.83 to 1.67 with a confidence interval range from 0.58 to 2.98. For the 3rd trimester of exposure results (OR) range from 0.88 to 2.00 with a confidence interval range from 0.62 to 3.62. These inconsistent results illustrate the lack of uniformity in the methods employed, difference between cross section, variability of variable's definition, and the lack of studies, particularly in Europe.
For studies focusing on the whole pregnancy for the relationship between pregnancy adverse outcomes risk and air pollutant exposition: NO 2 [ The Pedersen's study also had nearly significant results for NO 2 exposure associated with LBW (OR = 1.09; 1.00-1.19) and for PM 10 exposure associated with LBW (OR = 1.16; 1.00-1.35). Overall, the results reveal that the risk of adverse outcomes including: PTB [64,85,86,93], LBW [64,73,75], SGA [68,91] was not found to be significantly associated with any of the pollutants. As for the other windows of exposure (each pregnancy trimester), results are very heterogeneous and there appears to be no clear trend regardless of the model used. For NO 2 exposure results (OR) range from 0.81 to 1.28 with a confidence interval range from 0.91 to 1.74. For PM 10

Main Characteristics
The meta-analysis presented in this study was conducted for 2 combinations between one air pollutant and two birth outcomes during different windows of exposure, when at least four studies were available for the same combination. More precisely, the 2 combinations were NO 2 exposure and related with birth weight and PTB. Table 5 describes the measures of the association of the studies included in the meta-analysis.       In order to differentiate the health effect related to each trimester and entire pregnancy, stratified analyses have been performed, only when this is possible. For the combination between NO 2 and preterm birth, it was conducted for the entire pregnancy only. Following these conditions, we produced, finally, 5 meta-analyses. Of these, heterogeneity (Q-test) tests indicated one meta-analyses with high I 2 (I-square indicator) values (above or close to 50%) for which random effects models were applied (for the other four combinations, fixed models were used). Heterogeneity varied from 25.2% to 72.3%, indicating that measurement methods, sample properties, and characteristics varied both among and within different studies.

Birth Weight
As shown in Figure 5, the exposure of NO 2 during any windows of exposure on birth weight was not statistically significant. The overall analysis did not reveal a significant decrease of birth weight in pooled beta for any windows of exposure: for second trimester the pooled beta is: −8.35, 95% CI (−23.04, 6.34) (Figure 6), for the third trimester: pooled beta = −7.04, 95% CI (−19.90, 5.81) (Figure 7). It is interesting to note here that the exposure of NO 2 during the first trimester tends to show a nearly significant decrease of birth weight in pooled beta = −13.63, 95% CI (−28.03, 0.77). Finally, regarding whole pregnancy, as shown in the Figure 8, the exposure of NO 2 during the entire pregnancy on birth weight was not statistically significant. The overall analysis did not reveal a significant decrease of birth weight in pooled beta (fixed models: pooled beta = −1.40, 95% CI (−6.08, 3.29)).

Preterm Birth
As shown in Figure 9, the exposure of NO 2 during the entire pregnancy on birth weight was not statistically significant, and did not reveal a significant increase of the risk of preterm birth in pooled OR (pooled OR = 1.07, 95% CI (0.90, 1.28)).

Sensitivity Analysis
To estimate the stability of our results, sensitivity analysis was performed by recalculating the pooled effects estimates after omitting one study each time as long as there remained at least 4 studies (Appendix B). We found that the effect estimates of each 10 µg/m 3 increase in NO 2 exposure during the entire pregnancy on birth weight showed no significant change by removing one single study, suggesting that the combined results were relatively stable and reliable. This is except for the sensitivity analysis of the association between birth weight and NO 2 exposure during the third trimester of pregnancy, where the omission of the study of Clemente et al. (2016) [83] induced a reverse of the association that was hitherto negative (Table A1); however, the result was still not statistically significant (beta = 2.5, 95% CI = (−9.18, 14.30)). Small variations were visible, and while point combined estimates were rather similar, the precision level of the confidence interval decreased.       . Association between preterm birth and NO2 exposure during the entire pregnancy. Figure 9. Association between preterm birth and NO 2 exposure during the entire pregnancy.

Main Finding
Our systematic review does not show significant results, but despite this a trend is apparent in that NO 2 exposure during the whole pregnancy seems to increase the prevalence of LBW. In addition, the result of published European studies included in our systematic review tend to show an increased risk of LBW with a 10 µg/m 3 increase in PM 2.5 and PM 10 , specifically for long-term exposure including exposure during last trimester and whole pregnancy. By contrast, no significant excess risk of adverse birth outcomes has been found regardless of pollutant or short-term window of exposure (each trimester).
Our meta-analysis does not reveal a significant result, and the exposure of NO 2 during the first, second or third trimester on birth weight was not statistically significant. The overall analysis did not reveal a significant decrease of birth weight in pooled beta. For the PTB outcome and the exposure of NO 2 during the entire pregnancy, the overall analysis did not reveal a significant increase of the risk of preterm birth in pooled-OR.
The characteristics of the different studies (design, adjustment, definition of the outcomes ....) (see Appendix F, Tables A3 and A4) did not change the meta-risks estimated with the classical meta-analysis approach (data not shown).
These results for long-term exposure converge with international meta-analysis (see results in Appendix G) which show positive correlation between PM 2.5 , PM 10 , NO 2 , exposures during the entire pregnancy and LBW. [94] Conversely, international studies tend to show significant association between LBW and ambient air pollutant also during short-term exposure.
These results could be partially explained by methodological limitations inherent in the heterogeneity of the method of exposure assessment, definition of adverse birth outcome, definition of confounders and critical windows of exposure, thus limiting the number of studies usable in the meta-analysis which can reduce the statistical significance of possible risk.
The main hypotheses for the biological mechanism are that ambient air pollution could cause inflammation, oxidative stress, affect placental growth, decrease placental exchange, lead to endocrine disruption, etc. [95,96]. More specifically, oxidative stress induces DNA damage and mitochondrial DNA damage, and fosters inflammation, which appear to be important mechanisms of fetal growth [83,[97][98][99]. Another specific mechanism affects the placenta; the maternal and fetal circulation are separated by the placental barrier; this barrier contains placental transporters that can regulate or facilitate external compounds [100,101]. Transient receptor potential channels are highly expressed in the placenta, and they can be affected by air pollution exposure. Non-human animal studies reveal that these receptors play important roles in placental development and regulating the fetal-maternal interface in mice models [102].

Outcome Data: Case Selection
We identified many pathways whose outcome information can lead to a bias in the assessments of association. Firstly, outcome definition itself could constitute a source of uncertainty and lead to qualification bias. Many studies investigated birth weight [66][67][68]70,72,76,80,82,88,89] or gestational age [67], but most investigated specific pathological outcomes; first, several studies investigated LBW and subtypes (VLBW, ELBW): birth weight <2500 g International Classification of Diseases 10th Revision; ICD-10: P07.0-P07.1 [64,70,73,75,81,84], birth weight <3000 g [65], VLBW between 1500 g and 2500 g [83] and ELBW <1500 g [83]. Several studies investigated PTB and subtypes: birth occurring before the 37th week of pregnancy; ICD-10: P07.2-P07.3 [64,65,69,78,81,85,87,90,92,93] birth occurring between the 33th and the 37th week of pregnancy [79], birth occurring between the 22th and the 36th week of pregnancy [74], birth occurring between the 30th and the 37th week of pregnancy [77], birth occurring before the 33th week of pregnancy [79], birth occurring before the 30th week of pregnancy [77], birth occurring before the 24th week of pregnancy [87]. Last but not least, some studies investigated SGA: birth weight or length below the 10th percentile according to standard percentile charts for sex and gestational age in the population; ICD10 codes in medical records, O36.5, P05.0, P05.1 [68,70,75,82,91]. Databases were drawn mainly from birth certificate information and health database from hospital information systems while other from institutes of national health statistics and cohort database. In addition, the databases used to collect health data including maternal and newborn characteristics are another source of limitation. PTB and LBW were the most frequently investigated outcomes in included studies. This is an expected finding because, according to the WHO, these outcomes are technically simple parameters to monitor prenatal health in a population and have short-and long-term public health implications. Assessment of gestational duration was most often based on the date of last menstrual period, which could introduce misclassification with recall bias depending on postconceptional bleeding, but also, menstrual irregularities, or late access to prenatal care [70].

Confounding Factors
Our findings need to be interpreted with prudence due to weaknesses that could affect the significance of the associations and then the redaction of accurate conclusions. The different adjustment factors used in each study and the different sample size may lead to difficulties between studies comparisons.

Exposure Assessment
Our systematic review revealed that several approaches for exposure assessment during pregnancy were implemented, and this could induce misclassification of exposure. Some papers have used average from monitoring existing stations [64,65,74,77,81,84,87] or a monitoring station-based approach with an average from all monitoring stations [89,92]. The size of the study area and the number of monitoring stations vary between studies and this may increase the level of heterogeneity of air pollution measurement. The number of monitoring stations varied between a minimum of 1 [65,87] and a maximum of 53 [89]. Consequently, there is a risk of bias when a small number of monitoring stations cover a wide area. The weak spatial representativeness of exposure influences the assessment of the residential exposure of pregnant women. Moreover, collection of these data is often based on national air quality guidelines and legislation and thus may not be optimal in the assessment of exposure and use with health data.
Even if modelling is the gold standard for environmental and health impact assessment, some bias may exist. Overall modelling approaches did not consider residential and daily mobility of pregnant women across the study area and thus exposure misclassification may occur. Finally, environmental modelling can hardly be applied for outdoor and indoor pollution concurrently; notably because of a lack of information on the correlation of indoor and outdoor air pollution depends on geographical and meteorological conditions, building types and systems, and air exchange rates [103].

Critical Windows of Exposure
The definition of window of exposure could induce exposure misclassifications. In our systematic review, two main approaches define the window of exposure in order to investigate the relationship between birth outcomes and residential exposure: long-term exposure, and short-term exposure.

Assessment Approach and Mean Level of Exposure
The results found in the studies selected may vary according to mean level of exposure in each country, and particularly in each area of study. Our systematic review reveals that the risk of adverse birth outcome tends to be higher among study areas with low air pollutant average concentration. However, we highlight that these studies used mainly monitoring station. Some studies tried to estimate the discrepancy between results in the association between air pollution and birth outcome with different methods for estimating exposure [104,105]. They found that the level of NO 2 during pregnancy estimated by the nearest air quality monitoring station (AQMS) and by the temporally adjusted geostatistical model (TAG), tend to show the same associations [104,105]. For PM, the use of the nearest AQMS or dispersion models indicated consistent results both in terms of exposure estimates and association with birth weight [105]. Studies tend to show that AQMS and kriging rather predict the average level of pollutant in the urban area, whereas local patterns of variation and LUR might be the most robust methods to predict long-term exposure in complex areas [106]. In this way, pertinence of the method used for the exposure assessment mainly depends on the time-window length and endpoints considered, the spatio-temporal variability of the pollutants and the population's mobility [76].

Limitations and Risk Estimate of Birth Outcome
The features of the studies described above-such as study population, study design, sample size, the classification and definition of infant death, exposure assessment, difference between interquartile (IQR) used to assess the increase of exposure (Appendix H) and confounding factors-could all, independently or in combination, affect the quality of each study itself and, also, their comparison in our systematic review. Some factors may overestimate while other one may underestimate the risk of birth outcome.
The loss of precision inherent to such a general classification scheme (the definition of outcome and included all live birth) may reduce the likelihood of detecting an association between low birth weight and the study exposures. For instance, broad groupings of low birth weight into all LBW including term and preterm birth have also hampered the ability to examine associations for specific LBW by diluting relevant cases.
One source of such limitation lies in the databases. Using linked birth-hospital databases may reduce the likelihood of missing information, because it includes all birth information collected throughout birth, rather than only from institutes of national health statistics and cohort databases. Missing data, if not included, may yield the same effect, so that risk estimates of birth outcome, in particular, may be inaccurate.
In addition, the various confounding factors included in the individual studies make difficult the comparisons between studies. An absence of systematic adjustment for commonly known factors may affect the measure of association and thus the comparisons of all the risk estimates-for instance, folic acid supplementation, or information on dietary factors which are known to decrease the risk of birth outcome. These risk factors tend to vary across the unit of analysis and if they are coincident with the exposure measures, then these spatial confounders will bias the results of the study.
Exposure misclassification may occur where the birth certificate address does not reflect the mother's true residence during the relevant window of fetal development. To assign exposure, many studies used maternal address at delivery rather than address around conception and during each trimester. This can have a particular impact on studies exploring the risk of birth outcome. Misclassification of exposure may occur following changes in residence during the pregnancy. Some studies revealed that residential mobility among pregnant women between conception and delivery ranged from about 12% in the former to 32% according country. In addition, this residential mobility may vary according to certain individual and contextual characteristics such as age, race, socioeconomic status and other factors including socioeconomic characteristics. This means that the exposure misclassification error due to using delivery address might be greater among younger mothers than among older ones, a phenomenon that might result in confounding-because age is also associated with the risk of poor pregnancy outcome. Therefore, where authors have restricted their analysis to women who resided at the address noted on the medical record before delivery, a slight increase of risk estimate may be observed.
Finally, misclassification of exposure may result from the use of postcode, census block or city level to define the location of maternal residence. These spatial units might not be valid measures of exposure level because they vary considerably in size and are irregular in shape. Therefore, the larger the spatial unit, the more likely it is that bias will be introduced due to heterogeneity within these units, and ecological fallacy may result.

European Versus International Systematic Review: Comparison with Previous International Systematic Reviews
The limited systematic review of European studies may explain the result obtained. Appendix D summarizes the main characteristics and results of previous systematic review and meta-analysis studies selected to compare our results. Most of the earlier reviews were based on cohort design studies like our systematic review. However, all previous systematic review was based on mostly US studies. Similar methodological issues were identified by previous systematic review including outcome and difference in the characterization of exposure and outcome and control of confounding factors.
However, the European average concentration of air pollutants seems to be lower than international average concentrations, moreover the results found in the studies selected may vary according to mean level of exposure in each country, and particularly in each area study. Our results reveal that, in Europe-based studies, the risk of adverse birth outcome tends to be higher among study area with low air pollutant average concentration.
Our findings for long-term exposure converge with international meta-analysis (see results in Appendix D) which show a positive correlation between PM 2.5 , PM 10 , NO 2 , exposures during the entire pregnancy and LBW [94]. Conversely, international studies tend to show significant association between LBW and ambient air pollutant also during short-term exposure.

Limitations
To complement the limitations described earlier, both our systematic review and our meta-analysis, present their own strengths and limitations. First, our search could suffer from study selection biases. Non-English publications of relevant articles may have been ignored. Furthermore, we cannot exclude the possibility that our systematic review could be impacted by publication bias. Indeed, unpublished results (including grey literature and results not statistically significant, which are not available) may influence our meta-analysis findings towards the statistical significance of the risk estimates

Public Health Implication
To date the main inherent limitation of environmental health risk assessments is related to uncertainties of the assumptions made about the dose-response function. More particularly, the potential limitations of geographic extrapolation of the shape of the risk function may be less well-defined in some geographic areas with the lowest concentrations. In some studies, the authors used the exposure-response function from only one cohort US study [107] while the other one used meta-analysis as a source to estimate the burden [108]. To our knowledge no European meta-analysis permits us to provide a more appropriate source of risk function in order to perform HIAs in European countries with the lowest concentration levels. Thus, the burden could derive from a non-coherent shape risk function that carries larger uncertainties. Our meta-analysis results provide pooled-risk for 5 combinations of air pollutant and birth weight and PTB, which may provide a coherent exposure-response function for environmental health risk assessments in European countries.

Conclusions
In spite of the limited number of epidemiological studies selected in the present literature review, our finding suggests that an increase air pollution exposure during pregnancy might contribute to adverse birth outcomes, especially LBW. This body of evidence has limitations that impede the formulation of firm conclusions and so new well-focused European studies are called for.
Our findings need to be interpreted with prudence due to weaknesses that could affect the significance of the associations and hence the drawing of accurate conclusions. Further studies, well-focused on European countries, are called for to resolve these limitations; in particular, the definition of the exposure assessment, the critical windows of exposure and the different adverse birth outcomes, which could affect the strength of association. Future studies could be based on this analysis of limitations of the current body of research, which may provide inspiration for research agenda improvements.     )) is calculated for each study included in the meta-analysis.

Assessment of birthweight
Medical data and Self-report (0.5) Medical data and Self-report (0.5) valid database (1) valid database (1)