Association between exposure to particulate matter during pregnancy and birthweight: a systematic review and a meta-analysis of birth cohort studies

Studies of the associations between maternal exposure to particulate matter (PM) and risk of adverse effects on fetal growth are inconsistent and inconclusive. This question can be well answered by carefully designed birth cohort studies; however, so far the evidence from such studies has not come to the same conclusion. We sought to evaluate the association between maternal exposures to PM and low birthweight (LBW) enrolling 14 studies from 11 centers, and to explore the influence of trimester and exposure assessment methods on between-center heterogeneity in this association. Data were derived from PubMed, Embase, Google Scholar, CNKI, and WanFang database, references from relevant articles, and results from published studies until March 2017. Using a random-effects meta-analysis, we combined the coefficient and odds ratios (OR) of individual studies conducted among 14 birth cohort studies. Random-effect meta-analysis results suggested that a 17% and 6% increase in risk of LBW was relevant to a 10 μg/m3 rise in PM2.5 and PM10 exposure concentrations at the 3rd trimester (pooled odds ratios (OR), 1.17 and 1.06; 95% confidence interval (CI), 0.94–1.46 and 0.97-1.15, respectively), but the null value was included in our 95% CI. Our results showed that exposure to PM2.5 and PM10 during pregnancy has a positive relevance to LBW based on birth cohort studies. However, neither reached formal statistical significance. Negative impacts on outcomes of birth is implied by maternal exposure to PM. Further mechanistic researches are needed to explain the connection between PM pollution and LBW.

Birthweight is a characteristic indicator of prenatal growth, and infants with low birthweight (LBW) is one of the adverse pregnancy outcomes. Infants with LBW are at a greater risk of mortality and morbidity than infants with normal birthweight, as well as having health issues in childhood and beyond. These effects include asthma, hypertension and compromised cognitive ability [11][12][13] . Accordingly, exploring the risk factors for LBW, in order to reduce the occurrence of LBW, is extremely important to public health. The connection between heavy air pollutants, exposure during pregnancy, and LBW in recent years are investigated by numerous studies. Laurent et al. found that LBW was positively and significantly associated with the zone but not total fine PM [14] . Similarly, a national study conducted in Canada discovered that there was a steady indication of a dose-response association for NO 2 but no PM 2.5 impact on LBW [15] . Also, daily PM 2.5 with individual gestational ages of births in the contiguous United States was linked by another national study. This study indicated no overall significant positive connection between LBW and PM 2.5 exposure during pregnancy [16] .
A number of researches explored significant relationships between PM exposure and LBW [17][18] . These inconsistent and controversial results suggest that quantitatively integration and interpretation of available evidence produce more accurate results for policy decisions and clinical need is necessary.
Results from several independent studies can be quantitatively integrated using meta-analysis, a most commonly used statistical method [19] . The relationship between PM exposure of pregnant women and neonatus birthweight has also been quantitatively analyzed by a number of meta-analyses [6,[20][21] . However, all these meta-analyses observed notable heterogeneity among the studies included. Therefore, integrating various study results was necessary through a meta-analysis.
Here, we collected several studies that evaluated the effects of PM (PM 2.5 and PM 10 ) exposure during pregnancy on LBW, followed by employing a model of meta-analysis to estimate the effects of PM exposure on LBW during various pregnancy phases based upon birth cohort studies. The study site, sample, publish year, and exposure measurement methods were further evaluated for their potential influence on our meta-analysis.
Birth cohort studies provide the strongest evidence to comprehend the incidence and progression of diseases made possible by frequent follow-up data. None of the systematic reviews to date have paid special attention to the evidence from birth cohorts. Therefore, we present a systematic review and meta-analysis to estimate the effects of PM exposure during the different gestational periods with LBW.

Search strategy
We perform and report the corresponding results in this meta-analysis based on the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) guidelines [22] .
All publications indexed in English-language databases including PubMed, Google Scholar, and Embase; as well as Chinese-language databases such as China National Knowledge Infrastructure (CNKI), WanFang databases before March 6, 2017, were systematically searched for studies which can be included into our meta-analysis. A combination of the following keywords were used for searching the relevant literatures: ("air pollution" OR "particulate matter" OR "fine particulate matter" OR "fine particles" OR "PM" OR "PM 10 " OR "PM 2.5 ") AND ("cohort" OR "observational" OR "longitudinal" OR "follow-up") AND ("birth weight" OR "BW" OR "change in birth weight" OR "low birth weight" OR "LBW" OR "term low birth weight" OR "TLBW" OR "adverse birth outcomes" OR "adverse pregnancy outcomes"). Simultaneously, the references of relevant publications and meta-analysis were also investigated manually.

Selection criteria
Eligible studies included were considered if they satisfied the following conditions: (1) using cohort study design (e.g., not descriptive study, case-control design and experimental design, randomized controlled trial, etc.); (2) LBW was defined as a live birth weighing less than 2500 g, including term LBW (TLBW) and preterm LBW (PLBW); (3) sample size, partial regression coefficient (β) for birthweight, usable risk estimates (e.g., odds ratio (OR), risk ratio (or relative risk, RR) or necessary data for calculation) for LBW, and its 95% confidence intervals (CI), or necessary information from which these results could be inferred; (4) other risk factors that could impact the outcomes of pregnancy had to be modified, including but not limited to maternal age and infant sex. If a birth season, maternal tobacco, or alcohol consumption during pregnancy and socioeconomic status were also modified; (5) only publications in English or Chinese were considered.
In the last step, we excluded the birth records directly from the database or national public health system, and sources of PM pollution from indoor were also ignored.

Quality evaluation
Independently, JY and SF conducted a quality assessment of each study, included in our study, referring to the criteria derived from the Newcastle-Ottawa Scale (NOS) [23] . The NOS, which is used for assessing the quality of nonrandomized studies in metaanalyses, is available at: http://www.ohri.ca/programs/ clinical_epidemiology/oxford.asp. All literature included in this study were cohort studies. The items of NOS score for cohort studies was divided into three domains: selection of cohort (four points), comparability of the cohort (two points), and assessment of outcome (three points). The quality of the study was considered high or moderate if the sum score was eight points or greater or between five and seven points, respectively. This ensured that each of the eligible articles was of high or moderate quality.

Data extraction
Data was extracted from all eligible studies by two independent researchers (JY and SF) based on a standardized form. We resolved discrepancies by discussing with a third researcher (XB) extracting information consisting of the first author's name, year of publication, study location, study name, sample size, pollutant, exposure assessment, PM exposure windows (if a research implied connection between PM exposure during the whole pregnancy and/or trimester-specific periods and low birthweight, the assessments were obtained completely), outcome definition, covariates in the final model, and OR/RR/hazard ratio estimates with corresponding 95% CIs for all categories, continuous exposure of interest, or both, from each included research.
As gaseous pollutants were often different among studies, we obtained assessments from models of a single pollutant for results that covariates were fully modified. In this meta-analysis, we preferred the results that would depend on a larger number of pollutants.
For instance, we extracted the study of Michael Brauer et al. [24][25] over those from [26] for the results because the former study covered both PM 2.5 and PM 10 for the assessment method based on the monitoring network being more common across present studies. We preferentially chose this method to potentially reduce the heterogeneity among researches in this study.

Meta-analysis and statistical analysis
Various studies have been reported with different increments (e.g., increased with an interquartile range) or compared to a reference category. To pool estimates from the studies enrolled in, all risk estimates (OR) of PM 10 or PM 2.5, mass concentrations were converted to a uniform exposure of 10 mg/m 3 . Effect estimates were categorized by gestational period (whole pregnancy and trimester-specific). Weighting the inverse of the variance, we used a random-effects model to compute the pooled ORs and corresponding 95% CIs for each outcome of interest. Thereafter, the I 2 statistic test was assessed to evaluate heterogeneity among estimates from primary studies (50% or less for low-, 51%-75% for moderate-, and 76% or more for high-heterogeneity), respectively. We also performed a series of sensitivity analyses by removing study singly to examine whether the results were strongly influenced by a specific study. Finally, publication bias might have existed and was detected with conducting funnel plot asymmetry, and then we evaluated funnel plot symmetry through Egger's regression.
All two-sided tests were of α = 0.05. Statistically significant findings were considered as those with a pvalue < 0.05. We performed all statistical analyses with Stata version 11.0.

Search results and study characteristics
After a systematic search and review, we initially searched a total of 1,768 published English and Chinese literature; however, after excluding duplicates, reviews, and case-report articles, 1,650 kinds of literature remained. After glancing at these titles, 1,523 articles were further excluded as they were considered inappropriate regarding the interested endpoints or exposures. We reviewed the abstracts of the remaining 127 articles in detail. Sixty-two additional papers were removed since they focused on PCB/NO 2 exposure to indoor pollution, pollutants from the road, or mechanism researches, leaving 65 articles for an in-depth review. A flow chart of the selection is shown in Fig. 1. After carefully reviewing these articles, we distinguished fourteen for the final analysis from 2004 to 2016 [18,24,[27][28][29][30][31][32][33][34][35][36][37][38] , of which six studies assessed PM 2.5 and LBW, twelve studies assessed PM 10 and LBW, and three studies assessed both birthweight and LBW. The quality of included studies Table 3 presents the study-specific quality according to Newcastle-Ottawa quality scale.
Pooled estimate of the effect of the PM mass concentrations on low birth weight In Fig. 2 and Fig. 3, we estimated overall risks and the risk of LBW caused by PM 2.5 was 1.03 (pooled OR, 95% CI: 1.01-1.06) and pooled OR of PM 10 was 1.04 (95% CI: 1.00-1.07). Heterogeneity test indicated a moderate heterogeneity among six and twelve articles respectively, which was 59.2% for PM 2.5 -LBW (P = 0.002) and 50.8% for PM 10  Another subgroup analysis was carried out by the study location (Europe and America; and other regions), study samples (≥1,000, and < 1,000), published year      On the other hand, we collected articles which used birth data directly from the national birth registry or hospital-birth records to explore the connection between PM exposure during pregnancy and LBW; the results were displayed in Fig. 4 and Fig. 5. The pooled the estimate of PM 10 for the entire pregnancy (OR = 1.07, 95%:1.02, 1.11) was larger than other trimesters, although no statistical significance of the three estimates can be obtained. We also found that heterogeneity was the lowest for the 3rd trimester and the highest for the 1st trimester in Fig. 5.
According to Egger's tests, except for the P-value (P = 0.025) of PM 2.5 exposure in the 3rd trimester, no significant publication bias for the two pollutants can be seen in Table 5.

Discussion
Fourteen eligible studies were identified and collected in our meta-analysis, and the associations between the PM mass concentrations and the risk of LBW were quantitatively assessed based on birth cohort studies. Results suggested that maternal exposures to PM during the entire pregnancy and trimesters had a slight positive trend to associate with LBW, but the results were not statistically significant, which consisted well with the findings of the previous meta-analyses. Sapkota et al. reported that slight but formally non-statistically significant increases in risk of LBW was associated with the entire pregnancy PM 2.5 (summarized OR: 1.09; 95% CI: 0.90, 1.32) and PM 10 exposure (summarized OR: 1.02; 95% CI: 0.99, 1.05) [39] . Although Stieb et al. [6] reported that with a 10 mg/m 3 increase for PM 2.5 or a 20 mg/m 3 increase for PM 10 , the pooled odds ratios concerning entire pregnancy exposure of PM 2.5 -LBW and PM 10 -LBW were 1.05 (0.99-1.12) and 1.10 (1.05-1.15).
To the best of our knowledge, this is the first metaanalysis study so far reporting on the association between PM pollution and LBW based on birth cohort. All of the studies enrolled in this meta-analysis were birth cohort records and most of these studies did not show an association of statistical significance between PM exposure and LBW. We supposed the reason for the difference lies in that Stieb included the articles based on a national birth registry or state center databases. To verify our supposition, we utilized 21 studies which were directly from databases of various centers.   The in-depth evaluation of the evidence from birth cohorts is one of the main strengths of this review. If all articles were based on registered data, we could not get information comprehensively. A cohort study is an effective way to demonstrate the associations.
More or less, this meta-analysis has some limitations; although less heterogeneity in some subgroups, high or moderate heterogeneities appeared in many of the subgroup analyses. These findings illustrated that the heterogeneity may also be affected by other factors. The socioeconomic status were not investigated due to the limitation in quantity of relevant studies. Accordingly, further studies are warranted to examine the origins of heterogeneity as more meaningful studies are conducted in the future.