Study population
The design of the Japan Environment and Children’s Study (JECS) has been described previously in detail [10-12]. The direct web link to the JECS is https://www.env.go.jp/chemi/ceh/en/index.html This study followed the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) statement for observational studies. Briefly, pregnant women in Japan were recruited for the JECS between January 2011 and March 2014. Women who 1) lived in any of the Study Areas selected by the fifteen Regional Centers located in the country at the time of recruitment; 2) had an expected delivery date after August 1, 2011; and 3) were capable of understanding the Japanese language and completing a self-administered questionnaire were included in the study [10, 12].
The present study used the “jecs-ag-20180131” dataset, which was released in March 2018 and contains information on 104,065 fetal records (Fig. 1). Among women with multiple pregnancies during the study period, data for the second or third pregnancy was excluded (n =1,003); pregnancies with miscarriages, stillbirths, or missing data (n = 3,860) were also excluded. Overall, 99,202 pregnancies were included in the analysis.
The JECS protocol was approved by the Ministry of the Environment’s Institutional Review Board on Epidemiological Studies (no. 100910001) and by the Ethics Committees of all the participating institutions. Written informed consent was obtained from all the study participants.
Data collection
The study participants completed questionnaires throughout their pregnancies and postpartum periods; i.e., during the first and second/third trimesters (n = 92,550), and at one-, six- and twelve-months after delivery. The medical records at the time of registration and just after vaginal delivery or cesarean section were transcribed by doctors, research coordinators, nurses, or midwives.
Information regarding maternal or paternal demographic factors was obtained from the questionnaires completed during pregnancy. Postpartum information was collected from the questionnaires completed during the six months after delivery.
Outcomes, exposure, and covariates
The primary outcome was the occurrence of postpartum depression. We used the postpartum Edinburgh Postnatal Depression Scores (EPDS) at one and six months after delivery and the postpartum K6 scores within one year after delivery as the primary outcomes [13,14]. The EPDS is a validated, standardized questionnaire consisting of 10 screening items that is commonly used to identify a risk of perinatal or postpartum depression. As the cutoff value for the EPDS, we used a score of ≥9 as a positive result for postpartum depression [15,16].
The K6 self-administered questionnaires were assessed using a five-category scale (4 = all the time, 3 = most of the time, 2 = some of the time, 1 = a little of the time, 0 = none of the time), with possible scores ranging from 0–24. According to a Japanese validation study for the K6 questionnaire in the general population, the performance of the K6 questionnaire using an optimal cutoff of ≥13 to indicate severe psychological distress was excellent when the performance was examined using an area under the receiver operating characteristic curve (AUC), with values as high as 0.94 (95% confidence interval (CI) = 0.88 to 0.99) [17].
The participants were divided according to mode of delivery into three categories: vaginal delivery without analgesia; painless vaginal delivery, including epidural analgesia, spinal-epidural analgesia, or paracervical block; and cesarean section.
The covariates included maternal age (categorized as <20, 20–29, 30–39, ≥40 years) , maternal body mass index (BMI, categorised as <18.5, 18.5–25.0, ≥25.0 kg/m2), maternal educational status (categorised as junior high school or high school, higher professional school or professional school, junior college or college, postgraduate college), annual income (categorised as <200, 200–400, ≥400–600, ≥600–800, ≥800–1,000 JPY × 10,000; 1 USD = 103.5 JPY, December 2020), recurrent miscarriage (yes vs. no), mode of pregnancy (natural conception vs. others), parity (never vs. ≥once), drinking history (categorised as never, abstinence before pregnancy, abstinence from this pregnancy, continuance drinking), maternal smoking history (categorised as never, abstinence before pregnancy, abstinence from this pregnancy, continuance smoking 1 - 10 cigarettes per day, continuance smoking 11 - 20 cigarettes per day, continuance smoking over 21 cigarettes per day), pre-K6 during first trimester and second/third trimesters (categorised as mentioned above), marriage status at second/third trimester (categorised as married, non-married, divorced, partners’ death) and at six months after delivery (categorised as married, divorced, partners’ death, others), sex of child (categorised as male, female, unclear), Apgar scores at 1 and 5 min (<7 vs. ≥8), inborn errors of metabolism (categorised as nothing, require recheck, require complete check-up, confirm the diagnosis), neonatal anomalies (yes vs. no), breast- or bottle-feeding, frequency of infant crying (categorised as cry well and keep crying, sometimes but stop soon, not too much), and cooperation of partner with nurturing at one month (categorised as always, sometimes, very little, nothing) and at one year after delivery (categorised as nothing, very little, sometimes, well, very well).
Data analysis
The maternal and postpartum demographic characteristics of the participants were shown with the proportion for discrete data. The Fisher exact test was used to compare the association between the outcome and each variable. Binomial logistic regression analyses were performed by adding all the covariates to calculate the adjusted ORs (aORs) for the association between mode of delivery and postpartum depression. Since missing data can potentially undermine the scientific credibility of causal conclusions, we applied a multiple imputation method to reduce the potential non-response bias created by missing data and to improve the precision of the estimates when calculating the aORs [18,19]. A total of 20 models, in which all the available variables were used as predictors and outcomes, were created to estimate the aORs. To prevent multiple comparisons possibly yielding false-positive findings, we adopted the Benjamini-Hochberg method and assessed statistical significances by obtaining the q-values adjusted for false discovery rate. All the statistical analyses were performed using IBM SPSS Statistics for Windows, version 24.0 (IBM Corp., Japan).