Ethics approval

In this register-based retrospective study, all data were fully anonymized before being accessed for research purposes. All results were presented on an aggregated level, and no individual could be identified.

Under the Data Protection Act (Swe. lag (2018:218)), no consent to participate was required for analyzing this anonymized data. The Regional Ethical Review Board in Stockholm approved this study (2008/1182-31/4).

Study population

Our study population comprised 752,988 women with singleton pregnancies who had a live birth at or after 37 weeks and before 42 weeks gestational age between 1 January 2007 and 31 December 2014 in Sweden. For our analysis, we collected data from the Swedish Medical Birth Register (MBR), which contains the International Classification of Diseases, version 10 (ICD-10) codes of diagnoses during pregnancy and childbirth [15]. Using a unique personal identity number [16], we linked data from the Swedish Medical Birth Register to the National Patient Register and the Prescribed Drug Register for diagnoses of depression before and after childbirth. We also linked the data to the Cause of Death Register, the Education Register, and the Total Population Register for information of covariates and censoring [17–19].

We excluded 99,280 women with a history of depression before childbirth, classified as having an ICD-10 diagnosis code for depression (F32-F39) or postpartum depression (F53) from the National Patient Register (starting from 1 January 1973) or a filled prescription for an antidepressant from the Prescribed Drug Register (starting from 1 July 2005). As the birth cohort started in 2007, we obtained at least one year of data of prescribed dispensed drugs before pregnancy to capture information on prenatal depression. Because postpartum depression has been linked to preterm birth or ongoing illness of the baby [20,21], we restricted our study cohort to term births with a non-anomalous pregnancy. To reduce outcome misclassification, we only included women born in Sweden or another Nordic country, who may have fewer language or cultural barriers in seeking specialized care for depression [22]. After the exclusions, there were 486,722 women for analyses (Fig 1). In main analyses, we further excluded women with missing data (N = 38,862, 8.0%) which left 477,860 in the regression models.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Flowchart of the study population.

https://doi.org/10.1371/journal.pone.0255938.g001

Exposure variable

For our primary exposure of interest, we classified PPH by using an ICD-10 code (O72) during the birth hospitalization. This code was typically applied when the estimated blood loss is over 1000 ml, keeping with guidelines from the Swedish Society of Obstetrics and Gynaecology [23]. Because perioperative bleeding >1000 ml during cesarean birth was coded as hemorrhage (O67.8), we also included this code in our PPH classification. The ICD-10 codes for PPH have been validated in Sweden with high sensitivity (88.9%) and specificity (99.4%) [24].

Outcome variable

We classified PPD based on the presence of at least one of the following up to 12 months after birth: a filled outpatient prescription for an antidepressant or an ICD-10 code for depression (F32-F39) or postpartum depression (F53) associated with an inpatient hospitalization or outpatient care. Antidepressants were identified using the World Health Organization Anatomical Therapeutic Chemical (ATC) Classification codes beginning with N06A [25]. S1 Table lists all the associated ATC codes of antidepressants prescribed and dispensed at pharmacies under the categories of tricyclic antidepressants, selective serotonin reuptake inhibitors, monoamine oxidase inhibitors, reversible Inhibitor of monoamine oxidase-A, and other antidepressants.

Women were censored if they died or emigrated within 12 months of hospital discharge. Unless PPD was captured before the death, women who committed suicide were censored in the same way as other causes of death. The dates of death and emigration were collected from the Cause of Death Registry and the Total Population Registry, respectively.

Covariates

Based on literature review [3,26–28] and our expert opinion regarding clinical plausibility, the following covariates were selected as potential confounders: maternal age, parity, highest maternal education, cohabitation status (measured as "cohabiting" or "not-cohabiting" with a partner), maternal smoking status in early pregnancy (categorized into "no smoking", "one to nine cigarettes per day", or "ten or more cigarettes per day"), body mass index (BMI) recorded at the first antenatal visit, gestational age at birth, and birthweight. Data for all covariates (except education) were sourced from the Medical Birth Register. Data for each mother’s highest achieved education level were sourced from the Education Register.

Missing data

We identified 38,862 (8.0%) women with missing data for at least one covariate. Data for BMI, cohabitation status, and smoking were missing in 6.5%, 4.3%, and 3.9% of the study population. Mode of delivery, birthweight, and maternal age were missing in less than 0.2% of the study population. We considered these variables missing at random and performed complete case analyses for women who did not have missing data for any covariate. S2 Table presents the distribution of all variables from the complete case analysis and those excluded for missingness.

Statistical analysis

We used Cox proportional-hazards models to examine the association between PPH and PPD, adjusting for maternal and obstetric covariates. Kaplan-Meier survival functions were plotted to describe the association between PPH and PPD up to one year after giving birth. To explore whether the associations between PPH and PPD differ according to the mode of delivery, we performed stratified analyses within each of the following subpopulations: spontaneous vaginal birth, instrumental vaginal birth, planned cesarean birth, and emergency cesarean birth. To account for the potential effect of missing data, in secondary analysis, we performed multiple imputations using chained equations (MICE) with 50 imputations and replicated the main analysis on the complete population. Data management was performed with SAS^® version 9.4 (SAS Institute Inc, Cary, NC). All statistical analysis was performed with STATA^® version IC 16 (StataCorp, College Station, TX).

Sensitivity analysis

First, because the risk of PPD may vary according to the severity of blood loss, we reclassified our outcome as severe PPH. Severe PPH was classified by a transfusion ICD-10 procedure code (DR029 from the Medical Birth Register) associated with a PPH ICD-10 code. For this analysis, severe PPH was compared to no PPH as the reference group. Second, to account for clustering of births of the same woman, we restricted our sample to the first birth of the women in the study cohort.

Sensitivity analysis

In our first sensitivity analysis in which our outcome was severe PPH, a null association was observed between severe PPH (PPH with transfusion) and PPD (adjusted HR: 0.99, 95% CI 0.75, 1.29). Table 3 shows the association between severe PPH and PPD stratified by mode of delivery. Consistent with the main finding, there was no increased risk of PPD associated with severe PPH across all strata. The associations between PPH without transfusion and PPD were consistent with the main analyses, with the point estimation being the highest in instrumental vaginal birth but the confidence intervals overlapping null 1.25 (0.93, 1.67). In the second sensitivity analysis restricting the study cohort to the first deliveries only, a null association was also observed overall (adjusted HR = 1.08, 95% CI 0.98, 1.19) and across the strata by mode of delivery (S4 Table).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Adjusted Hazard Ratios for postpartum depression after exposure to postpartum hemorrhage classified by with and without blood transfusion.

https://doi.org/10.1371/journal.pone.0255938.t003

Principal findings

In this population-wide study of 486,722 term pregnancies in Sweden, we observed that PPH was not associated with PPD. Results of our sensitivity and stratified analyses were similar to those of our main analysis, which provide additional evidence to substantiate the robustness of our main findings.

Study strengths

A key strength is that the large size of a national sample of all women who gave birth in Sweden between 2007 and 2014. Linked data from Swedish datasets also allowed us to follow up for 12 months after giving birth and account for a detailed set of pre-pregnancy and pregnancy-related confounders, which is clinically relevant and adding to previous knowledge. In addition, our dataset did not contain any missing data for our primary exposure (PPH) or outcome (PPD).

Study limitations

Because the accuracy of PPD diagnoses using ICD-10 codes in Sweden is unknown, it is unclear whether PPD was over- or under-ascertained. However, in Sweden, universal screening for PPD with the Edinburgh Postnatal Depression Scale (EPDS) [29] is provided to mothers at 6–8 weeks postpartum during child health service [30]. To confirm our findings, future studies may use EPDS data to assess the accuracy of PPD diagnostic coding with ICD-10 codes.

We attempted to account for depression that existed before childbirth. But we could not identify women with antepartum depression that did not have an ICD-10 code for depression and did not receive antidepressants during pregnancy. Therefore, some PPD cases may have had an antepartum depression, and the observed association was confounded by pre-existing depression. Also, excluding women with a history of depression limited the generalizability of our findings.

Besides depressive symptoms, PPH may also induce emotional trauma and posttraumatic stress disorder [31]. However, few studies have investigated this link [32]. Future population-based studies may need to examine if PPH increases risks of other psychiatric morbidities such as posttraumatic stress disorder or anxiety.

Although the accuracy of ICD-10 coding for PPH is high [24], validation studies rely on blood loss measurements for determining PPH coding accuracy. It is well recognized that blood loss estimation can be inaccurate, with blood loss often being underestimated in severe cases [33] and in caesarean births [34]. Therefore, the prevalence of PPH may have been underestimated, which may have biased point estimates in our regression analyses toward the null. Of note, the PPH diagnosis was given at >1000 ml estimated blood loss in Sweden during the study period, which is higher than the commonly used threshold of 500 ml [35]. Our null finding with the higher threshold also suggested a lack of substantial impact of PPH on PPD. Besides PPH, postpartum anemia could not be accurately measured given the data, leaving the association between postpartum anemia and PPD unclear. Thus, future studies with appropriate data may help elucidate the link between postpartum anemia and PPD.

To avoid confounding by indication, we excluded women with a known history of depression. This likely explains why the PPD prevalence was lower than the PPD prevalence reported in overall birthing populations [3].

Interpretation

Based on our findings, PPH was not associated with an increased risk of PPD. In particular, whether PPH was associated with increased risk of PPD after instrumental vaginal birth remains inconclusive. These findings require validation in large scale prospective studies with a more accurate measure of PPD. If confirmed, women who experience PPH may not require additional surveillance for PPD after childbirth.

Compared to our findings, previous studies examining the association between PPH and PPD produced inconsistent results. This may be partly because these studies did not account for confounding from pre-existing depression or antidepressant exposure [11–14]. Furthermore, definitions for PPH and PPD vary across these studies. A qualitative study by Sentilhes et al. suggested that women who had a PPH requiring arterial embolization had long-term psychological trauma, including anxiety and depression [13]. In contrast, an Australian clinical based cohort study on women who had PPH of over 1500 ml or a low postpartum hemoglobin concentration [14] concluded that the physical and emotional well-being of the cohort did not differ from women in a general postpartum population, mirroring the findings of our study. Similarly, a Swedish population-based survey examining 446 women from Uppsala county [11] showed no association between PPH over 1000 ml and PPD, defined as having the EPDS score ≥12.

Compared to these previous studies, our study used a large sample from the national population. We deliberately focused on Nordic-born women who were more similar in accessing medical care for PPD, who had no pre-existing depression, and who gave term births. By restricting on the confounders and factors associated with measurement errors, we were better able to disentangle the association between PPH and PPD risk.

Other pre-pregnancy, antepartum, and postpartum social and psychosocial factors may be more predictive for PPD, in contrast to a single risk factor of PPH. Future studies are warranted to explore the interplay of obstetrical, social, and psychosocial factors in affecting PPD risk. It is also unclear whether a similar association would be observed in countries with a differing prevalence of PPH or PPD, in addition to different social, psychosocial factors, and obstetric practices. Therefore, future studies are needed to find out if the lack of a strong association between PPD and PPD is generalizable in other countries.