Assessment of the association between ambient air pollution and stillbirth in the UK: Results from a secondary analysis of the MiNESS case–control study

We examined whether the risk of stillbirth was related to ambient air pollution in a UK population.

to identify and address modifiable factors that reduce stillbirth have been a healthcare and research priority for the last decade. 4,5ir pollution is associated with a range of health problems and may be a modifiable factor for stillbirth if a causal link can be established.Recognised toxic air pollutants include NO and NO 2 (known collectively as NO x due to their ability rapidly to inter-covert) and particulate matter (PM).In the UK, around 2.2 million tonnes of NO 2 are produced annually: one-quarter from power stations, half from fuel combustion in road transport, and the rest from other industrial and domestic combustion. 6imilar processes are responsible for PM emissions, with ~190 000 tons of PM 10 and 100 000 tons of PM 2.5 produced per year. 7urrent evidence linking air pollutants to stillbirth is inconsistent.In an early systematic review of 13 studies, Siddika et al. 8 found increased NO 2 exposure was associated with increased stillbirth risk in the prenatal period.Cohort studies in California (mean NO 2 = 68.7 microg/ m 3 ) and Wuhan, China (mean NO 2 = 58.8microg/m 3 ) respectively found associations with NO 2 exposure during the first 9 or third trimester and the risk of stillbirth. 10onversely, other studies found no effect of NO 2 exposure on pregnancy outcome, [11][12][13] though these studies noted lower levels of mean NO 2 ranging from 35.1 microg/m 3 (Czech Republic) to 41.1 microg/m 3 (London, UK).Similar variation was seen in the effects of PM 2.5 exposure; one study using personal air monitoring in 362 women found exposure throughout the pregnancy (mean PM 2.5 43 microg/m 3 ) reduced birthweight. 14A population cohort study from 2006 to 2010 in Ohio found stillbirth (mean PM 2.5 13.3 microg/m 3 ) was more frequent with high PM 2.5 levels (median + interquartile range) in the third trimester, 15 but other studies have shown no such association. 9,11he most recently published meta-analysis found maternal third trimester PM 2.5 exposure (per 10 microg/m 3 increment) increased the odds of stillbirth (pooled odds ratios [OR) 1.094, 95% CI 1.008-1.180)and the entire pregnancy exposure was also associated with stillbirth (pooled OR 1.103, 95% CI 1.074-1.131). 16Notably, there were high levels of heterogeneity (for example, I 2 values for PM 2.5 ranged from 62.1% to 88.7%), indicating a high proportion of variability explained by differences between the included trials. 16As well as differences in exposure between studies, the lack of consistency in findings may result from variation in methods of measuring air pollution exposure which do not always accurately capture the complex spatial and temporal patterns. 17Furthermore, many studies at the population level are not able to appropriately identify and adjust for potential confounding factors including social deprivation, maternal age, smoking status and maternal body mass index (BMI). 18evertheless, a relation between air pollution is biologically plausible.In high-resource settings the most common association with stillbirth is being small for gestational age (SGA), 19 which is thought to be secondary to placental dysfunction; abnormalities of the placenta and cord are the most frequent abnormality found in stillbirths in high-resource settings. 20Particulate matter, consistent with particles derived from air pollution, have been found in the placenta and umbilical cord blood and their presence is associated with placental nitrative stress. 21,224][25] Thus, air pollution could be linked to stillbirth via placental dysfunction, leading to an SGA fetus.
Therefore, this study aimed to estimate the effect of exposure to NO 2 , NO x and PM 2.5 in pregnancy and the periconceptual period on stillbirth risk accounting for potential confounding factors.

| Study population and ethics
This study was a secondary analysis of the dataset from the Midlands and North of England Stillbirth Study (MiNESS), a case control study of late stillbirth (≥28 weeks' gestation). 26Women were recruited between April 2014 and March 2016; participants were pregnant between August 2013 and October 2016.Ethical approval for the study was obtained from the UK Health Research Authority (Reference: 13/NW/0874).Cases were singleton pregnancies that ended in a late stillbirth where the fetus did not have major congenital anomalies.Controls were women with on-going pregnancies matched to the expected gestational age profile of cases in an approximate 2 : 1 ratio.For cases, potential participants were approached either after the stillbirth of their infant or during their pregnancy.For both cases and controls, the participants' postcode, demographic and clinical characteristics were recorded using an interviewer-administered questionnaire. 27Detailed characteristics of the study population have been described previously. 26Study participants were excluded from this analysis if there was missing information about maternal BMI, smoking status or age, women had pre-existing medical conditions known to be associated with increased risk of stillbirth, maternal residential postcode was missing or not correctly recorded or the maternal residential postcode was >20 km from a stationary monitoring site.
Index of Multiple Deprivation (IMD) values, based on the 2015 scores, were obtained for each participant included in the study based on maternal residential postcode at the time of birth.The IMD decile for each postcode was obtained from the Indices of Deprivation explorer: http:// dclga pps.commu nities.gov.uk/ imd/ idmap.html.Infants were categorised as small for gestational age (SGA) if their birthweight was <10th centile based on customised weight centiles. 28The presence of placental or cord insertion abnormalities was assessed at the time of birth by the midwife -placentas were recorded as being abnormal if there was evidence of retroplacental clot, widespread calcification or there was abnormal placental shape or marginal cord insertion.

| Participant and stationary monitoring site mapping
Maternal residential postcodes for each participant at the time of birth were converted to Ordnance Survey grid references X (easting) and Y (northing).The residential location for each participant was mapped in ARCMAP by Arc Geographical Information System (GIS) version 10.4.1.The locations of all stationary monitoring sites in England (excluding roadside monitoring stations) were plotted alongside the residential converted postcodes.Roadside air pollution monitoring stations were not included as they capture roadside pollution levels that are unlikely to represent concentrations of pollutants in wider areas.Monitoring stations that did not measure air pollutants over the time periods of the pregnancies included in this cohort were also excluded, leaving 24 monitoring stations in this analysis.Participants were assigned to the nearest monitoring station using 'Near Analysis', which is a geospatial analysis tool in ArcGIS.Buffering was then used to identify participants who were more than 20 km from the nearest monitoring station for exclusion from further analysis.This was a pragmatic threshold chosen to balance inclusion of participants to give an adequate sample size against inaccuracies in estimation of pollutants as distance from a monitoring station increased; a threshold of 20 km has previously been applied to evaluation of the association between developmental anomaly and air pollution. 29

| Air pollution exposure estimation
To estimate air pollution exposure levels for the participants, background maps were generated using the Pollution Climate Mapping (PCM) model.PCM models are generated each year under the Department for the Environment, Food and Rural Affairs (DEFRA) Modelling of Ambient Air Quality (MAAQ) contract. 30The methodology used to create these maps is explained elsewhere; 31 the modelled maps are available from the UK-AIR Modelling Data page (https:// uk-air.defra.gov.uk/ data/ gis-mappi ng/ ).The annual average air pollution data for 2013-2016 was used, as this was the time period during which the participants' pregnancies occurred.Ordinary Kriging, a method of stochastic modelling, was applied to the air pollution data using tools available in the ArcGIS software toolbox; this method was chosen as it provides the most unbiased estimates of personal exposure compared with measurement of personal exposure. 32Once the surface air pollution maps had been created, they were converted to a raster to allow further spatial analysis using ArcGIS.Annual air pollution exposure estimates (NO 2 , NO x and PM 2.5 ) for each participant were extracted, using the 'extract value to point' tool in the ArcGIS software.Annual estimates were then extracted from GIS for each participant.To increase the temporal accuracy of air pollution exposure, as part of the spatio-temporal (S-T) exposure model, MAFs (Monthly Adjustment Factors) were applied to the PCM annual mean concentrations. 32MAFs were calculated for each pollution monitoring station by dividing measured monthly mean pollution concentration by the annual mean pollution concentration at that site.Each MAF was applied to the PCM data to calculate the monthly exposure estimates during each participant's pregnancy, including the 3 months' preconception.

| Air pollution exposure periods
To investigate potential differential effects of air pollution at different stages of pregnancy, exposure estimates were calculated for the 12 weeks prior to conception, the first 16 weeks of gestation, from 16 weeks until delivery, and over the entire pregnancy from conception to delivery; these epochs were chosen to represent the preconception period, early and late pregnancy.Each participant's temporally adjusted air pollution estimates were used to calculate the average exposure over each time frame for each participant by adding the estimated pollution for that period and dividing it by the number of months in that time frame.If an exposure period included partial months (i.e.occupied only part of a month) then the average exposure/day was calculated and added to the overall exposure.The length of the second half of pregnancy and the whole pregnancy exposure varied depending on the gestational length of the pregnancy.

| Statistical analysis
Statistical analyses were performed using SPSS (IBM, Version 27) and STATA (Version 14; StataCorp).Descriptive statistics were used to describe the study population and the fetal characteristics.They were presented as categorical variables with frequency (N) and percentage (%), comparisons between cases and controls were assessed using chi-square tests.Adjusted standardised residuals were used to interpret the results when a significant association between variables was seen.Pearson's correlation was used to assess the correlation between the air pollutants.An independent sample t-test was used to determine the difference between the mean air pollution exposure estimates for the control and case groups.Generalised linear regression models were then used to assess the relation between air pollution exposure, IMD and household income.
The association between air pollution exposure and stillbirth risk was assessed using univariable and multivariable logistic regression.Multivariable logistic regression adjusted for household income, maternal BMI, maternal smoking, IMD quintile and household smoking and parity, as these covariates are associated with risk of stillbirth in this and other study populations. 26,33,34Regression models were used to calculate the effect of the levels of each air pollutant in microg/m 3 increments on the risk of stillbirth in each exposure period.A P-value <0.05 was considered statistically significant.A mediation model was used to investigate the mechanism of the observed effect of air pollution on stillbirth risk both directly and indirectly via birthweight centile.

| Demographics of study participants
The total population of the MiNESS dataset, prior to exclusion criteria being applied for this study, comprised 291 stillbirths and 733 livebirths.This analysis included data from 835 participants, 238 cases (stillbirths) and 597 controls (see Figure 1).Participants' data were excluded for missing information about: maternal BMI, smoking status or age (n = 50); pre-existing medical conditions known to be associated with increased risk of stillbirth (n = 12), maternal residential postcode missing or incorrectly recorded (n = 32) and maternal residential postcode not being within 20 km of a stationary monitoring site (n = 95) (Figure 1).The study population remained similar to the total MiNESS population (Table S1).Table 1 shows maternal and infant characteristics of the study population.In this study population, maternal smoking, parity, maternal BMI and the size of the infant were all found to have a significant relation with stillbirth risk.

| Air pollution exposure estimates for cohort
NO 2 and NO x exposure levels were strongly correlated (r = 0.95) but there was only a moderate correlation between PM 2.5 and NO 2 and NO x (r = 0.58 and r = 0.62, respectively).Calculated exposures of NO 2 and NO x varied widely, whereas exposure to PM 2.5 was more consistent across the cohort (Table S2).

| Spatial variation in air pollution
Figure 2 shows the spatial variation of NO x , NO 2 and PM 2.5 for mean monthly pollutant concentrations in the study period.Participant locations included in the final data analysis are shown on the maps.During the study period, levels of NO x and NO 2 were significantly higher in Birmingham and Manchester than in Leeds, whereas levels of PM 2.5 were significantly higher in Birmingham than in Leeds and Manchester (P < 0.001: Figure 3).

| Air pollution and stillbirth risk models
Table 2 presents the unadjusted and adjusted odds ratios (ORs) and 95% confidence intervals for the risk of stillbirth in relation to known associations and levels of NO x , NO 2 and PM 2.5 in single pollutant models at different stages of pregnancy.Univariable analysis demonstrated that NO 2 was associated with risk of stillbirth in the periconception period but not at other time points during pregnancy, but there was no association between stillbirth and NO x or PM 2.5 exposure preconception or during pregnancy.Figure 4 shows the air pollution exposure at each stage of pregnancy for case and control participants.The only difference in air pollution exposure between the case and control groups was higher exposure to NO 2 in the 3 months prior to conception in women whose pregnancies ended in stillbirth (P = 0.003).
Univariable analysis also found that maternal smoking and household smoking (where mother and partner smoked), household income <£20k, primiparity, presence of placental or cord abnormality, SGA and increasing BMI were all associated with stillbirth and these were included in the multivariable analysis with prenatal air pollution data.

| Relation between air pollution exposure and deprivation
The relation between IMD and household income with pollution exposure was explored using regression analysis (Figure S1A-C).There was a significant relation between IMD and all pollutants, but no continuous relation with selfreported household income.Women living in more deprived areas were more likely to live in areas with higher levels of pollution.For each improvement in IMD decile, pollution changed by −0.74 microg/m 3 (95% CI −0.85 to −0.62, P < 0001), −1.18 microg/m 3 (95% CI −1.37 to −0.99, P < 0.0001) and −0.12 microg/m 3 (95% CI −0.15 to −0.07, P < 0.0001) for NO 2 , NO x and PM 2.5 respectively.
After adjustment, low household income and deprivation were not significantly associated with stillbirth, but the remaining covariates, including preconceptual NO 2 exposure, remained statistically significant.Each 10 microg/m 3 increase in prenatal NO 2 exposure was associated with a 42% increase in the risk of stillbirth (OR 1.42, 95% CI 1.12-1.80).
Within the cohort, 50.8% of cases of stillbirth were SGA, and being SGA was strongly associated with stillbirth.We examined whether there was a relation between higher levels of air pollution and SGA but none was found.

| DISCUS SION
Our analysis of a large granular dataset derived from a stillbirth case-control study demonstrated that pregnancy exposure to UK ambient air pollution during pregnancy is not associated with a significantly increased risk of stillbirth.Our results concur with findings from systematic reviews 8,16 and large cohort studies 35 that failed to demonstrate significant independent relations between NO 2 or NO x exposure and stillbirth.These more recent findings contrast with earlier studies that demonstrated an effect, which may be the result of overall reduction in ambient air pollution over time. 36In our study, median levels of all pollutants examined were at or below recommended safe annual limits for exposure (NO x 50 microg/m 3 , NO 2 40 microg/m 3 , PM 2.5 35 microg/m 3 ). 37Ambient air pollution effects on birth outcomes in the UK were most recently examined in the UK in London between 2006 and 2010. 11,38In that study, NO 2 pollution levels were significantly higher than those observed in our study (mean NO 2 41 microg/m 3 , NO x 73 microg/m 3 and PM 2.5 23 microg/ m 3 ), with more significant effects on birthweight occurring at NO 2 levels >50 microg/m 3 . 38This further supports the hypothesis that ambient air pollution effects are marginal at the levels now found in much of the UK.

| Strengths and limitations
Our study used a contemporary prospective case-control dataset in its design, which enabled the independent effects of different exposures to be examined between stillbirths and controls.As the study took place over 2 years, this design allowed for us to control for the potential confounder of seasonal differences in pollution exposure as well as established risk factors for late stillbirth.The MiNESS study used face-to-face data collection, which means potential errors in exposure estimation introduced by subjects moving area during the index pregnancy are reduced.Another strength of our study is its applicability to contemporary western populations.Air pollution in the UK has fallen since 1992, when annual mean urban NO 2 concentrations were 60 microg/m 3 , to ~20 microg/m 3 in 2018. 39This means that studies examining older cohorts have limited applicability to current exposures.However, due to the relatively low level of air pollution exposure in the UK MiNESS population it was not possible to examine the potential detrimental effects of ambient pollution exposure above recommended safe limits on stillbirth risk, which limits the applicability of our findings to pregnant women in similar developed country environments.This may also explain why the previously observed relation with air pollution in later pregnancy seen in cohorts with much higher overall exposure was not demonstrated here.
Another limitation of our study is that, in common with the majority of population level studies, we have had to estimate exposure to air pollution using exposure models derived from air quality monitoring stations.This approach is problematic when estimating the pattern of distribution of non-gaseous pollutants, which are more variable over distance.Thus, judging the effects of exposure to these pollutants at levels typically seen in the UK requires prospective individual exposure measurements.However, a prospective study design would not be possible when studying rare outcomes such as stillbirth.Thus, studies must try to balance the need for adequate sample size against the need for granular data to quantify exposure to air pollution and other potential confounding exposures.It is important to note that although we have adjusted for potential confounders in our dataset, we have not been able to adjust for unmeasured confounders.

| Context of study findings
We demonstrated a statistically significant association between preconceptual NO 2 exposure and stillbirth that was independent of confounders, but as there was no similar association with NO x exposure or effects in the first trimester, this may not be a true biological effect.Nevertheless, a recent large cohort study based in Utah examining the incidence of physician-diagnosed FGR found a similar effect of preconceptual NO 2 exposure on fetal growth restriction (FGR). 402][43] As FGR, GDM and HDP are related to stillbirth, the observed association could be related to  several causal pathways originating in the periconceptual period.Within the MiNESS cohort, 50% of stillbirth cases were SGA.Our exploratory sub-analysis on this group did not find a relation between air pollution levels and SGA; this may mean that placental damage in some cases of stillbirth is not associated with reduced fetal growth, but more likely demonstrates the problem with using SGA as a surrogate for FGR, which can occur in fetuses above the 10th centile that have failed to reach their genetic growth potential.However, this can only be diagnosed by observing a decline in fetal growth velocity, which is not measured routinely in the UK.Therefore, it is not possible to determine whether the non-SGA cases in the MiNESS study were affected by FGR.However, along with the finding of a much higher rate of abnormal placental morphology in the cases of stillbirth, our findings suggest much higher rates of placental dysfunction and FGR-mediated stillbirth than identified by SGA alone.Note: Adjusted odds ratios (OR) derived from multivariable analysis.The adjusted odds ratios for NO x , NO 2 and for PM 2.5 are per 1 microg/m 3 .Statistically significant ORs shown in bold.For maternal factors, n is given as number of participants in each category.For the air pollution data, case and control average exposure levels are given.
T A B L E 2 (Continued) The estimated air pollution exposure levels for the case and control participants prior to conception, in early pregnancy, in late pregnancy and from conception till birth.(A) Levels of NO x , (B) levels of NO 2 and (C) levels of PM 2.5 .There was a significant difference in NO 2 exposure prior to conception between pregnancies that ended stillbirth and live birth, with the stillbirths being exposed to significantly more NO 2 during this period.During the other time periods there was found to be no significant difference in NO 2 exposure between the case and control participants.No significant differences were found at any time period for NO x or PM 2.5 exposure.Error Bars: ± 2 SD.**P = 0.003.
Prenatal exposure to NO 2 has been associated with low birthweight in large cohort studies. 38This effect may be through direct effects of NO 2 that has entered the maternal circulation or from a systemic inflammatory response induced by contact with NO 2 in the maternal lungs. 44Either of these responses may induce oxidative stress and DNA damage, which in turn can lead to impaired placental function, 45 most likely through either genomic or mitochondrial DNA damage. 46,47Why the early developing placenta could be more susceptible to NO 2 -mediated damage than the placenta later in pregnancy is unclear, but it may be related to a more established ability of the placenta to withstand insults or reduced maternal inflammatory responses as pregnancy progresses.
Our data support existing data that deprived communities are exposed to higher levels of air pollution, which may compound other social determinants of health, leading to higher rates of stillbirth. 48Data from a detailed cohort in Los Angeles identified that high levels of perceived social stress, already associated with stillbirth in the MiNESS population, 34 exacerbated the effects of air pollution on birthweight. 49Furthermore, air pollution may modify the effects of cigarette smoking on birthweight. 50Therefore, attempts to improve air quality in women who reside in the most-deprived areas may mitigate some of the increased risk of adverse outcomes experienced by these women.

| CONCLUSION
We have demonstrated that in a contemporary UK setting where ambient pollution exposures were beneath recommended levels, exposure to air pollutants during pregnancy does not appear to be associated with an increase in the risk of stillbirth.We observed a small effect of periconceptual NO 2 exposure which was associated with increased stillbirth risk, suggesting a possible period of increased susceptibility.Therefore, although it is clearly desirable to reduce ambient air pollution to improve population health, it is unlikely that policies to achieve this will significantly affect overall stillbirth rates in the UK and similar settings.

AU T HOR C ON T R I BU T ION S
AEPH and JMDT contributed to all aspects of the study design and obtained funding.AEPH had overall responsibility for the study.LH, JEM, EDJ and JMDT analysed the data with input from AEPH and AP.All authors were responsible for the drafting of the paper.All authors gave approval for the final version of the paper.

F I G U R E 1
Flow diagram showing eligible participant identification from MiNESS population.

F I G U R E 2
Heat map showing variation in levels of air pollution in England in 2015.(A) Variation in NO x , (B) variation in NO 2 and (C) variation in PM 2.5 .Red dots indicate the maternal residence of case participants and green dots show residential location of control participants.Two maps are shown per air pollutant, one of which shows the spatial distribution of the pollution, while the other shows the location of the case and control participants across the regions.Key shows the levels of air pollution at each colour in the in microg/m 3 .

F I G U R E 3
The log estimated pollution exposure levels for the participants based on location of birth.The exposure levels between participants with births in the Birmingham area, Leeds area and the Manchester area were examined.There was found to be significant geographical variation in pollution exposure for all of the pollutants.Error bars: ± 2 SD.*P < 0.001.T A B L E 2 Maternal factors and air pollution and their relationship to stillbirth.
T A B L E 1SGA, small for gestational age.