In this comprehensive quantitative analysis, we summarized the most up-to-date evidence of the associations between prenatal and postnatal exposure to ambient and household air pollution on SGA, IUGR and HAZ from 31 epidemiologic studies published on or before May 2019. The pooled effect estimates consistently suggest a moderate, but noteworthy increase in the risk of SGA (including IUGR). A 10 µg/m3 increase in PM2.5 during the first trimester was associated with a 2.3% increase (95% CI: 0.03–4.5%) in risk of SGA, and between a 2% and 11% elevated risk is observed with increased exposure to PM during various gestational exposure periods, including the first and second trimesters (PM2.5) and over the entire pregnancy (PM10). The direction of the associations between household air pollution and moderate and severe stunting (defined as HAZ < − 2 SD) are consistent with that for ambient air pollution and SGA, though the associations are not statistically significant.
We identified three previous systematic reviews linking air pollution and SGA, IUGR, and HAZ [13, 14, 38]. Two reviews focused on the impact of ambient PM2.5 exposure on various adverse pregnancy outcomes that included SGA. Only one review provided pooled effect estimates , while the other offered qualitative observations of the evidence . In a 2015 review, Zhu et al.  reported ORs of 1.07 (95% CI: 1.05–1.10) and 1.06 (95% CI: 1.02–1.10) for SGA per increase in PM2.5 exposure during first and second trimesters, as well as statistically significant associations for the third trimester and the entire pregnancy. Our pooled OR estimates of 1.02 and 1.04 for SGA per 10 µg/m3 increase in PM2.5 during the first and second trimesters, respectively, are slightly smaller in magnitude than those from previous meta-analyses. One explanation for this discrepancy is that individual epidemiologic studies that were published after the 2015 meta-analysis tend to report effect estimates that are lower in magnitude and statistical significance as compared to those from earlier studies. Several of the newly included studies containing PM2.5 were from cities in China where much higher PM2.5 levels (e.g. 38–71 µg/m3) were reported [36, 39]. This is in sharp contrast to earlier studies that were conducted in cities with much lower pollution levels (e.g., 4–14 µg/m3).
This phenomena of decreasing effect size with increasing PM2.5 concentrations has been previously documented , which indicates a non-linear association between concentration and response. Nonetheless, our findings are supported by existing literature on the impact of ambient air pollution on stunting defined as HAZ. A recent study found that prenatal exposure to the 1997 Indonesian forest fires is associated with a 0.41 in HAZ (or 3.4 cm) at age 17, which implies a loss of 4% of average monthly wages for approximately 1 million Indonesian workers born during this period . Another study reported that in Bangladesh, where stunting prevalence is as high as 36%, children with a high quartile of PM2.5 exposure (52–73 µg/m3) had 1.13 times the risk of stunting (HAZ < − 2) than that of children in the lowest quartile of exposure .
For household air pollution, we identified five epidemiologic studies for moderate and severe stunting defined by HAZ, including three studies published after a previous meta-analysis . Bruce et al.  reported a pooled OR of 1.27 (95% CI: 1.12–1.43) for moderate and severe stunting associated with household air pollution from cooking with solid fuels. Similar to the pooled estimates for ambient air pollution, our pooled effect estimates for household air pollution from cooking with solid fuels are smaller, but nonetheless in the same direction of association. The volume of evidence from both ambient and household particulate exposure suggests a high level of confidence regarding causality between ambient particulate pollution and prenatal determinants of stunting.
There is a strong biological basis for the relationship between air pollution exposure and low birthweight. Kannan et al.  reviewed the evidence from existing literature and determined there are potentially five distinct mechanisms at work: oxidative stress, inflammation, coagulation, endothelial function and hemodynamic responses. While precise biological mechanisms connecting air pollution with impaired fetal growth are unknown, it is commonly hypothesized that transplacental and postnatal exposure to particulate matter may result in oxidative stress leading to DNA damage. Induced acute placental and pulmonary inflammation, increased possibility for coagulation, and triggered endothelial dysfunction are also hypothesized biological mechanisms.
Our meta-analysis had several limitations. First, we observed a moderate to high degree of heterogeneity across gestational exposure periods and exposure metrics. Such heterogeneity may be explained by differences in study design methods and exposure assessment, covariate adjustment, study population and/or PM chemical composition that varies by study location. Further studies are needed to explore the independent and joint effects of early life exposures to air pollution and nutrition, and the effect of PM constituents on stunting-related risks. Second, we did not evaluate studies with other exposure periods (e.g., months) due to scarcity of relevant studies. Hence our observed effects on stunting-related risks based on the entire pregnancy period and specific trimester periods may not infer biological significance as for timing over gestation. Moreover, while majority of the included studies evaluated stunting-related outcomes as categorical measures (e.g., HAZ < − 2 SD), it is important to note that growth faltering is a gradient, and children above the conventional cut-off points for SGA/IUGR/HAZ may still experience suboptimal linear growth, especially in low-resource settings [7, 12], and thus the actual impact of air pollution on stunting or growth failure may be underestimated. Last but not least, we did not assess the quality of each included study.
These limitations are balanced by the substantial strengths of our study. While much of the material in this review has been published previously, additional value derives from updating existing reviews (seven versus 26 for ambient PM pollution in the current review, three versus nine for household air pollution) using comparable methods, including effects from both the continuous and categorical measures of PM2.5 and PM10 exposures. Prior to this review, an effort to join all available childhood stunting-related outcomes, as well as both the ambient particulate pollution exposure and household air pollution from solid fuel, had not been attempted. Our pooled estimates were robust in sensitivity analyses, as demonstrated by using stricter definitions of stunting, or removing one study from the main analysis. There was also no significant publication bias according to Begg’s and Egger’s tests.