The association between postpartum depression and air pollution during pregnancy and postpartum period: a national population study in Taiwan

Epidemiological evidence shows the association between air pollutants and several mental health outcomes, including depression, sleep disturbance, dementia, childhood neurodevelopment and suicide. Pregnant women are believed to be more susceptible and vulnerable to environmental pollutants, and postpartum depression (PPD) is a prevalent debilitating mental disorder. However, data on the effects of exposure to air pollution during pregnancy and postpartum period on the risk of PPD remain limited. This study aimed to evaluate the association between exposure to ambient air pollution during pregnancy and postpartum period and the incidence of PPD. The Taiwan Birth Cohort Study recruited representative 12% of all newborn in 2005 and their mothers by two-stage stratified sampling, including 21 248 mother–infant pairs. The occurrence of PPD was assessed by a self-reported questionnaire. Exposure to air pollutants during pregnancy and postpartum period was estimated using hybrid kriging/land-use regression (LUR) and integrated LUR-machine learning model based on data from the air monitoring stations. Logistic regression was then conducted to determine adjusted odds ratios (aORs) of PPD in relation to air pollutants. A total of 21 188 women were included in the final analysis, among whom 3,648 (17.2%) developed PPD within 6 months postpartum. The occurrence of PPD was significantly related to exposure to ambient concentrations of nitrogen dioxide (NO2) during first trimester after adjustment [aOR: 1.081 per interquartile range (10.67 ppb), 95% confidence interval: 1.003, 1.165], but not to particulate matter ⩽2.5 µm in diameter or carbon monoxide. Exposure to ambient NO2 during early pregnancy was significantly related to the occurrence of PPD among the women investigated in this population-based study.


Introduction
Postpartum depression (PPD) is a common and serious mental health problem among women who have given birth. Despite variations in definition among population studies, estimates of prevalence in different countries range from 13% to 19% (O'Hara and McCabe 2013). PPD impairs maternal social and occupational functioning, and distresses the whole family. Furthermore, mother-infant interaction can be affected, resulting in numerous negative consequences on the physical health and cognitive, linguistic, emotional, social, and behavioral development of the offspring (Boath et al 1998, McMahon et al 2006, Norhayati et al 2015, Slomian et al 2019. The biological etiology underlying PPD is not fully understood. Most scientists believe that perinatal hormone changes may be involved, including steroid hormone, glucocorticoid, oxytocin, etc (Schiller et al 2015, Brummelte andGalea 2016). A larger number of studies exist investigating neuroendocrine changes, neuroinflammation, neurotransmitter alterations, circuit dysfunction, and also the involvement of genetics and epigenetics in the potential pathophysiological mechanisms contributing to PPD (Payne and Maguire 2019). These diverse mechanisms are also highly interrelated, and make the intricate etiology of PPD challenging.
Ambient air pollution has been reported to increase the risk of central nervous system disease and neuropsychiatric disorders, including depression, in animal and human studies (Block and Calderón-Garcidueñas 2009, Fonken et al 2011, Hahad et al 2020. A meta-analysis reported an increased risk of depression as a result of long-term exposure to particulate matter ⩽2.5 µm in diameter (PM 2.5 ) and shortterm exposure to particulate matter ⩽10 µm in diameter (PM 10 ), nitrogen dioxide (NO 2 ), sulfur dioxide (SO 2 ), and carbon monoxide (CO) (Zeng et al 2019). Another systematic review included studies published till August 2019 found only short-time NO 2 exposure had significant result, but not PM 2.5 , PM 10 or SO 2 (Fan et al 2020). In addition, a review took publications before September 2020 into consideration and found both long-term of PM 2.5 and PM 10 , and also short-term of PM 2.5 exposure increased the risk of depression (Liu et al 2021). Therefore, these may indicate that short-term of PM 2.5 , NO 2 and CO, as well as long-term of particulate matter air pollution exposure have adverse effects on mental health.
With the dynamic physiologic alterations in their bodies (Soma-Pillay et al 2016), women in pregnancy and in postpartum period are vulnerable to neuropsychiatric disorders, including PPD. However, to our best knowledge, only two studies have examined the relationships between PPD and air pollutants. A U.S. pregnancy cohort study found that higher PM 2.5 exposure in the second trimester was associated with increased total Edinburgh Postnatal Depression Scale (EPDS) scores and depressive symptom subscale scores among Black women (Sheffield et al 2018). A study conducted in Mexico revealed that an increase in average PM 2.5 exposure during pregnancy was associated with an increased risk of PPD at 6 months (Niedzwiecki et al 2020). In these two studies, only PM 2.5 was considered, but not other air pollutants.
Therefore, we assume that exposure to air pollution may contribute to PPD, although there is not much research to draw the conclusion. To exam the hypothesis, we used NO 2 and CO as surrogates for traffic-related air pollutants (TRAP) (Health Effects Institute 2010). The other important category of air pollutants are particulates. PM 2.5 is from multiple sources, has been documented to cause oxidative stress, systemic inflammation, and multiple health outcomes (Araujo 2011, Lu et al 2015. This current study examined a representative sample of pregnant mothers who gave birth in 2005, and investigated the relationship between PPD occurrence and maternal exposure to PM 2.5 , NO 2 , and CO during pregnancy and postpartum period, to determine the potential role of air pollutants in maternal PPD.

Study population and sampling
The Taiwan Birth Cohort Study (TBCS) is a national prospective longitudinal cohort study. A two-stage stratified random sampling method was performed on the Taiwan National Birth Registration data from 2005 to identify representative mother-infant pairs involved in the TBCS. First, 12 strata were defined according to administrative division (four strata: district, city, urban township, and rural township) and total fertility rate (three strata: low, medium, and high). All 369 townships in Taiwan were included, from which 89 were randomly selected as the primary sampling units. Second, newborns were proportionally selected from each primary sampling unit based on the principle of probability proportional to size. A total of 24 200 samples were obtained from the birth registry, which accounted for approximately 12% of all deliveries in Taiwan in 2005. Interviewers visited the selected candidates at their residence, and informed consent was obtained from the parents or main caregivers of the infant.

Questionnaire and dependent outcome
Structured questionnaires were administered by approximately 80 well-trained interviewers when the children were aged 6 months. Information regarding maternal conditions during pregnancy, infant birth outcomes, infant health conditions, parental socioeconomic status, and other demographics were obtained. Those who were unable to complete the questionnaire were excluded from the cohort. A total of 21 248 mother-infant pairs were included. In the survey, mothers were asked if they had experienced PPD. Answering 'yes' to this question was considered a positive response to the dependent variable. Mothers with missing or unsure answers to the question regarding PPD (n = 43) were excluded in the data selection process. In addition, whether or not the mothers have the disability identification of chronic mental health conditions based on Physically and Mentally Disabled Citizens Protection Act in Taiwan, or who met the criteria for a severe illness of chronic mental disorders as defined by the Taiwan's National Health Insurance (NHI) Administration, were further excluded (n = 17). Therefore, 21 188 participants included in the final analysis (figure 1).

Exposure assessment
In this investigation, PPD was compared to ambient exposure to PM 2.5 , NO 2 , and CO. Participants' exposed levels to PM 2.5 and NO 2 were estimated based on a hybrid kriging/land-use regression (LUR) approach (Wu et al 2018, Chen et al 2020. By integrating kriging-interpolated air pollution estimations with LUR models including culture-specific sources as potential predictors, this hybrid approach combines the strengths of both Kriging and LUR methods and represents a high percentage of explanatory power with R 2 of 0.88 for PM 2.5 and 0.90 for NO 2 , respectively. In addition, LUR-based extreme gradient boosting algorithm model was conducted to the prediction of spatial-temporal distribution of CO, and the R 2 value was 0.85 (Wong et al 2021). Each participant's home address was geocoded to the township level and linked to the spatial mean of township to determine their exposure to air pollution. Exposure periods were defined as (a) first trimester (from conception to 12th week of pregnancy), (b) second trimester (13th to 26th week of pregnancy), (c) third trimester (from 27th week of pregnancy to childbirth), and (d) postpartum 3 months (childbirth to 3 months postpartum). The date of conception was estimated by subtracting the infant's gestational age from the birth date, and the gestational age was determined based on the mother's self-response in the first-wave of cohort study. Mothers got the information of gestational age with ultrasound performed by obstetricians during the prenatal health check-ups.

Statistical analysis
The binary outcome variable was the occurrence of PPD, and the major predictor variables were the estimated exposure levels to ambient air pollutants including PM 2.5 , CO and NO 2 . Potential covariates were compared between mothers with and without PPD, including the sex, birth order, gestational age, birth weight of infant, tocolysis during pregnancy, delivery method, delivery season, breastfeeding, infant health condition, maternal age at childbirth, maternal educational level, prepregnancy body mass index, perinatal smoking or passive smoking history, perinatal alcohol consumption, marital status, urban living, annual household income, and family support. The urban residence was defined from seven levels of townships, with level 1 as the most urban and level 7 as the least, developed by Academia Sinica, Taiwan (Hou et al 2008). We categorized the townships in levels 1-2 as urban and those in levels 3-7 as rural. Still, ambient temperature in the same period as the air pollutant was further adjusted, considering the synergic association between temperature and air pollutants (Kalisa et al 2018) and also the potential health effect of temperature on depression . Ambient temperature was estimated using an ordinary kriging method, which had been described in detail (Shih et al 2020). Data on these other potential confounders were collected during face-to-face interviews for the survey, performed when the infant was aged 6 months. The chi-square test was used to compare categorical data and the t test was used to compare continuous data. Significant covariates were used to adjust further analyses for confounding factors. Logistic regression was employed for each air pollutant in each period of pregnancy (first, second, or third trimester) and the first 3 months postpartum adjusting for confounders. The findings were reported as the odds ratio (OR) and 95% confidence intervals (95% CIs) of the occurrence of PPD for each 10 µg m −3 increments in PM 2.5 , 0.1 parts per million (ppm) in CO, 1 part per billion (ppb) in NO 2 , as well as per interquartile range increase in pollutant concentration. Further sensitivity analysis was performed using two-pollutant models. Statistical analyses were performed using the statistical software package JMP version 14.0 (SAS Institute Inc., Cary, NC, USA), and statistical significance was set at P < 0.05 based on a two-sided calculation.

Results
In the 21 188 mothers included in the final analysis, 3,648 (17.2%) reportedly experienced PPD within 6 months postpartum. The demographic characteristics of the participants are summarized in table 1. Slightly more than half of the infants were boys, and no significant relationship was observed between the sex of infant and PPD. The following factors were significant covariates and were included in the further regression model, namely, first-born infants, tocolysis during pregnancy, cesarean section, premature or low-birthweight infants, delivery season, breastfeeding, poor health status in infant, maternal age, high education level, perinatal smoking or passive smoking exposure, perinatal alcohol consumption, other than married marital status, urban living, annual income of more than US$ 20 000 per year, and poor family support. Table 2 presents the distribution of exposure to each air pollutant during pregnancy and postpartum period. In 2005, the air quality standards in Taiwan had not regulated the level of PM 2.5 , but the concentration of PM 2.5 in our population was higher than WHO air quality guidelines of 10 µg m −3 annual average (World Health Organization 2006). The NO 2 and CO concentrations were in compliance with Taiwan's 50 ppm and 9 ppm regulatory standards at that time. Table 3 details the correlations among these air pollutants by different periods. The level of CO strongly correlated with that of NO 2 . Table 4 presents the results of the logistic regression in determining the relationship between PPD and maternal exposure to air pollution according to different periods. Exposure to NO 2 during first trimester was determined to be associated with PPD after adjusting for birth order of infant, tocolysis during pregnancy, delivery method, delivery season, premature and lowbirthweight infant, breastfeeding, children general health status, mother's age at childbirth, mother's education level, maternal perinatal smoking, passive smoking and alcohol consumption, annual household income, marital relationship, urban living and family supports. Exposure to PM 2.5 or CO was not related to PPD. After adjusting for ambient temperature, and also using two-pollutant models with PM 2.5 , average ambient NO 2 level during first-trimester (aOR: 1.081 per 10.67 ppb, 95% CI: 1.003, 1.165) and within the first 3 months postpartum (aOR: 1.078 per 10.48 ppb, 95% CI: 1.002, 1.160) were significantly related to PPD. However, if we put these two periods together to compete for effects, only first-trimester NO 2 remained statistically significant.
The final regression model including NO 2 exposure during first trimester was showed in table 5. The following factors were associated with an increased risk of PPD, namely, first-born infants, tocolysis during pregnancy, cesarean section, delivery season in spring, breastfeeding, poor health status in infant, maternal age between 25 and 35, high education level, perinatal smoking, passive smoking exposure, perinatal alcohol consumption, other than married marital status, and poor family support. However, there was no significant results from the effect of premature or low-birthweight infant, urban living, ambient temperature, nor household income.

Discussion
To our knowledge, this is the first study employing a nationwide representative survey that has detected an association between NO 2 exposure during pregnancy and the risk of PPD. Exposure to interquartile range, namely, 10.67 ppb higher in NO 2 in the first trimester was associated with an 8% increase in the likelihood of PPD.
The main source of NO 2 was traffic emission, thus it is viewed as the index of TRAP. These results accord with previous studies. A meta-analysis of seven studies revealed a significant association between short-term NO 2 exposure and depression in general pollution (OR: 1.02, 95% CI: 1.00, 1.04 per 10 mg m −3 increase, I 2 : 65.4%) (Zeng et al 2019, Fan et al 2020. Regarding the long-term effect, Spanish and Korean studies reported that depressive disorder was related to NO 2 exposure (Vert et al 2017, Shin et al 2018. However, a cohort study did not detect significant long-term effects of NO 2 exposure in Germany, Norway, and Finland (Zijlema et al 2016). The difference may result from lower exposure to NO 2 in European countries. These abovementioned studies have mainly investigated non-pregnant populations. One study conducted in Shanghai studied pregnant women and showed a significant relationship between NO 2 exposure and maternal stress levels (Lin et al 2017). Therefore, our study is the first one clearly pointing out that TRAP exposure, especially NO 2 , during early pregnancy would increase the risk of postpartum mental illness in women.
In this investigation, PM 2.5 was found unrelated to PPD occurrence, which was different from the conclusion of a previous study (Niedzwiecki et al 2020), despite the higher concentration of exposure in our study. The hybrid kriging/LUR model used in our study showed high explanatory power with R 2 of 0.88 and root mean square error of 7.86 for PM 2.5 (Wu et al 2018), suggesting reliable estimation of ambient PM 2.5 . The possible explanation for the nonsignificant relationship may be caused by different components of PM 2.5 between countries. Particulate matter is emitted from a variety of source type, such as fuels combustion, industrial processes, road dust and sea salt or soil erosion. The concentration of PM could not present the actual activities in the areas, which would result in the discordant finding. Additionally, we did not consider the long-term effects of particulate matter on depression as in previous studies (Liu et al 2021). Due to the small number of studies and inconsistent relationship between ambient PM 2.5 exposure and PPD, further research is warranted.
CO exposure was not statistically significant in the study. Previous systematic review reported a pooled estimate from three related surveys indicated that short-term exposure to CO was borderline associated with an increased likelihood of depression (OR = 1.01; 95% CI: 1.00, 1.01 per 0.1 ppm increase, I 2 : 70.2%) (Zeng et al 2019). Korean studies also reported long-term effect of CO exposure on depressive disorder (Shin et al 2018). Though CO was also viewed as one of the indexes of TRAP, we inferred that the non-significant results in our study may be due to the relatively low ambient CO concentration.
PPD was inquired in 6 month postpartum mothers in TBCS in this investigation. The criteria used by psychiatrists in Taiwan   depressive disorder (MDE) with postpartum onset was recognized in the DSM-IV-TR as PPD. Postpartum onset was defined as being within the 4 weeks following delivery. The essential feature of MDE is a period of at least 2 weeks during which an individual experience either a depressed mood or a loss of interest or pleasure in nearly all activities. Taiwan's NHI program covers a postpartum checkup 4-6 weeks after childbirth, as well as two well-baby clinic visits before the age of 6 months in Taiwan. Therefore, accessibility to medical care was adequate for the mothers if PPD were to develop during the first 6 months after childbirth. The prevalence of PPD in 6 month postpartum mothers was observed to be approximately 17.2% in our study. A 10.3% prevalence of PPD 6 weeks postpartum in Taiwan had been reported in 2001-2002 among 203 subjects by psychiatric specialists using the DSM-IV criteria (Teng et al 2005). Another prospective longitudinal study detected that 9.4% (22 out of 234) of mothers displayed postpartum MDEs in the 4th week postpartum . These two investigations were conducted by experienced psychiatric teams. However, they suffered from high rates of refusal and/or low followup. It was possible that women with more severe PPD were less able to complete the survey, leading to underestimation of the prevalence. Investigations using self-report structured questionnaires, mainly the Taiwanese version of the EPDS, revealed a prevalence of 20% or higher (Heh 2001, Huang andMathers 2001), which was slightly higher than the observed PPD in this current study. For the validation of the participating women's PPD status, reported PPD was compared with their answers to the 36-Item Short Form Survey (SF-36), Taiwan Version (Lu 2003), which was included in TBCS survey at 6 months after childbirth. The four subscales of SF-36, i.e. vitality, social functioning, roleemotional and mental health were compared between mothers with and without PPD (supplementary table 1 (available online at stacks.iop.org/ERL/16/084021/ mmedia)). Those mothers with reported PPD had lowered scores in the four subscales. This provided support for validity of the mothers' reported PPD.
In this study, PPD was found related to primiparas, tocolysis, cesarean section, delivery season, breastfeeding, poor infant health, maternal age at childbirth, high education level, perinatal smoking, passive smoking, alcohol consumption, other than married marital status, and poor family support. Most of these factors represent stressors for mothers (Akincigil et al 2010, Xie et al 2010. In Taiwanese, parents are more concerned about their first baby because of the numerous uncertainties in bearing a child, which causes psychological stress. A Japanese survey also reported higher depressive symptoms for primiparas (Iwata et al 2016). Furthermore, tocolysis reduces oxytocin (Ikeda et al 1984, Vrachnis et al 2011, and a relationship has been observed between low levels of oxytocin during pregnancy and PPD (Skrundz et al 2011, Moura et al 2016. The relationship between delivery season and PPD can be attributed to day length (Goyal et al 2018). The association between breastfeeding and PPD has been studied by several investigators. However, the causal direction and type of this relationship remain unclear (Pope and Mazmanian 2016). In our study, having breastfed was associated with an increased risk of PPD. Although breastfeeding might have certain protective effects against stress (Heinrichs et al 2002), the rates of mothers who breastfed or attitudes toward breastfeeding may differ between cultures and countries. Difficulties in breastfeeding may be a factor (Watkins et al 2011). Therefore, investigating the underlying mechanisms of breastfeeding would be valuable. Nicotine uptake from smoking may enhance the hypothalamic-pituitary-adrenal (HPA) axis response to stress, which increases the risk of depression development (Yu et al 2010, Reuter et al 2012. Furthermore, studies had noted maternal age was associated with the risk of PPD, (Aasheim et al 2012, Silverman et al 2017. Our findings differed from a study that reported that lower maternal education levels were related to higher risks of PPD (Goyal et al 2010). Studies in Korean and Chinese participants found that the association of depressive symptoms and educational status was complex, and the results may differ in distinct sociocultural situations (Jo et al 2011, Gan et al 2012. Therefore, various factors affecting the risk of PPD that may be interrelated were closely bound up, and further research is warranted to clarify the intricacies of PPD development.
There are two potential mechanisms of how air pollution leads to PPD. First, literature has demonstrated endogenous nitric oxide (NO) would be associated with alterations in the HPA axis hormones (Mancuso et al 2010). During pregnancy, the endogenous NO metabolite level is increased (Owusu Darkwa et al 2018). Then maternal NO metabolite levels should decline to normal level after giving birth (Okutomi et al 1997). Studies have demonstrated that the pathogenesis of PPD is be related to the change of dysregulation of the maternal HPA axis (Jolley et al 2007, Glynn et al 2013. Therefore, it is possible that exposure to high levels of ambient NO 2 during pregnancy may cause relatively larger-scale disruption of HPA axis activity after childbirth. The effect may cause extended the postpartum HPA refractory period, thus increasing the risk of PPD. The other potential mechanism is that air pollution affects the central nerve system through several cellular and molecular pathways, which can engender diseases caused by neuroinflammation, disruption of the blood-brain barrier, oxidative stress, endothelial dysfunction, and neuronal damage (Calderón-Garcidueñas et al 2008, Block and Calderón-Garcidueñas 2009, Brockmeyer and D'Angiulli 2016. These neuropathophysiological effects would contribute to PPD (Payne and Maguire 2019). Air pollution may induce dopaminergic neurotoxicity, resulting in depression symptoms (Levesque et al 2011, Yolton et al 2019 in both laboratory animal and human studies. Therefore, it was possible that a delay effect or accumulative effect of exposure to air pollutants in early pregnancy may induce PPD occurrence because of neuroinflammation or neurotoxicity. Our study has some limitations. First, PPD was based on a self-reported answer, and it was not a clinical diagnosis assessed by professionals in this questionnaire-based study. However, among those women with reported PPD, poorer score in the psychometric scores were found by SF-36. Therefore, such an assessment was still credible. Infant's gestational age was also an important information in our study, and it was reported by mothers at 6 months postpartum. In Taiwan, there were 99% of mothers giving birth in hospitals or clinics by obstetricians during recent 15 years. Before delivery, the gestational age would be measured by ultrasound during prenatal health check-ups. We therefore believed the bias would be accepted. Second, the calculated exposure to ambient air pollutants was not personal exposure, because indoor air pollutant levels and time-activity patterns were not considered. Applying personal environmental monitoring in such a large sample was impractical, especially among pregnant subjects. Still, earlier exposure to ambient exposure before pregnancy did not considered in our study, because of lacking previous living address data. Third, only the administrative district of each participant was provided because of ethical privacy considerations. Therefore, exposure was measured at the township level, which inevitably resulted in misclassification of exposure. Such misclassification may theoretically reduce the observed association. However, we observed a significant relationship between air exposure and the occurrence of PPD, which suggests that the actual effect may be stronger. Fourth, approximately 12.2% of mothers did not join the survey and the reason for rejection was not provided. Therefore, we do not know whether the rejection was related to PPD or not. Because the participation rate was relatively high, the influence of bias may be acceptable (Galea and Tracy 2007). Mothers with PPD may have been unable to join the study, which should not bias the exposure-outcome relationship, but could have caused misclassification, thus reducing the observed relationship toward the null hypothesis.
Our study has several strengths. First, the participants were representative of Taiwanese babies born in 2005 from a population-based cohort design, which minimized the uncertainty of random error when collecting detailed data on covariates. Second, the collection of exposure to air pollutants from the Taiwan Environmental Protection Administration was comprehensive and independent of the methods used to obtain the main outcome in our study. Third, the high-accessibility and low-copayment conditions of the NHI program allow people to habitually seek perinatal medical services, which should be associated with higher reliability of the dependent outcome when we approached mothers at approximately 6 months postpartum. Therefore, the assessment of PPD in our study was still reliable. Fourth, two major types of air pollutants, namely, TRAP and particulates were studied during different periods of pregnancy, which allowed for detection of potential effects from various pollutants during critical windows.
In conclusion, this investigation found that the occurrence of PPD was positively associated with maternal NO 2 exposure during early pregnancy in this representative birth cohort from 2005. The present study was the first to identify the effect of TRAP during pregnancy on maternal PPD. Further confirmation of the effects of these air pollutants and investigation of the relevant mechanisms are warranted.

Data availability statement
All data that support the findings of this study are included within the article (and any supplementary files).

Acknowledgments
This study was based on data from the TBCS Database, provided by the Health Promotion Administration, Ministry of Health and Welfare, Taiwan, R O C. We thank all children and parents who participated in this study, the interviewers who supported data collection, and all of the study groups who participated in the TBCS. We also thank the Department of Medical Research at National Taiwan University Hospital for helpful discussions during manuscript preparation.

Funding information
This study was supported by grants from National Taiwan University Hospital, Taiwan (Grant# 110-N4846) and Ministry of Science and Technology (Grant# MOST109-2621-M-002-021). The views expressed herein are the authors' own.

Ethical statement
This study protocol has been reviewed and approved by the Institutional Review Board of National Taiwan University Hospital (#202007068RINC).