Lifetime Prevalence and Risk Factors for Perinatal Depression in a Large Cohort of Women with Depression

Amongst women with a history of depression, this study sought to identify risk factors associated with reporting perinatal depression (PND)). Lifetime prevalence, length and severity of PND were evaluated, as well as the effect of PND onset either after previous depression episodes, or as the rst episode of depression. Reproductive measures included age at menarche, parity, age at rst birth, presence and severity of nausea and vomiting during pregnancy (NVP), and previous diagnosis of gestational diabetes, endometriosis or polycystic ovarian syndrome. Antidepressant measures included the number of antidepressants that had been tried, their ecacy and any side-effects. More details of these measures are provided in Supplementary Methods, which also lists the questions used to assess each characteristic.


Introduction
Background Perinatal depression (PND), including both antenatal and postpartum depression, commonly classi ed as a subtype of major depressive disorder (MDD) 1 , carries serious risk for both mother and infant. An estimated 53% of women with postpartum depression have "high suicidality" 2 , whilst the rate of selfharming thoughts is three times that of the postpartum community population 3 . Estimated economic costs of PND in the UK, of which 72% are for ongoing care of the child 4 re ect ndings that children of women with persistent and severe PND are at increased risk of adverse outcomes 1,5 .
The diagnostic criteria for MDD and PND are the same 6 , but the strongest known PND risk factor is a previous diagnosis of any psychiatric disorder 1,7-10 , not only MDD 11 . Other risk factors may also increase PND vulnerability. Possible psychosocial factors include stress and history of abuse 12 whilst biological factors include changes that accompany pregnancy, such as hormonal uctuations and increased in ammation 13,14 .
The complexity of these risk factors contribute to ongoing debate about the heterogeneous nature of PND in relation to MDD; in particular, whether it is simply another episode of MDD that happens to coincide with the perinatal period 15 ; or a subset of MDD, termed "reproductive depression", stimulated at times of hormone uctuation such as pre-menstruation, peripartum and menopause 16,17 ; or a distinct disorder, stimulated by changes occurring during pregnancy and con ned to the perinatal period 13 . One suggestion is that PND is itself heterogenous 10, 18 , with clinical subtypes differentiated by timing and severity of symptoms, perinatal complications, and history of psychiatric disorders. Silverman et al. 7 examined PND heterogeneity according to previous psychiatric disorders, and suggested two PND pathways: as a further episode of MDD occurring peripartum; or, alternatively, through perinatal complications leading to disengagement from the infant. However, a comprehensive investigation of the characteristics of women with PND, with and without a prior psychiatric history, has not been attempted.

Objectives
Using the Australian Genetics of Depression Study (AGDS), a large cohort study with over 20,000 participants self-reporting a depression diagnosis 19 , we examined PND heterogeneity, based on the presence or absence of previous major depression history. We sought to address two questions: 1) What are the differences in clinical and psychosocial characteristics between women with and without PND after a depressive episode prior to their rst pregnancy?
2) What are the differences in clinical and psychosocial characteristics between women whose rst episode of depression was during the perinatal period and parous women who have also experienced depression, but never during any perinatal period?

Study Design
Within a case cohort study of the etiology of depression, two groups of PND cases and two comparison groups of NPD cases were identi ed according to their history of prior depression. For both PND groups, the length and severity of their "worst case" of PND was measured. To investigate risk factors associated with PND after a previous history of psychiatric disorders, or as rst onset depression, both PND groups were compared with their comparison group of NPD cases, across a range of variables.

Setting: The Australian Genetics of Depression Study
The AGDS is a large ongoing case cohort study of the etiology of depression that recruited 20,689 participants (aged between 28 and 58 years; 75% female) during 2016-2018. The analyses conducted here are from participants enrolled prior to the initial data freeze in September 2018. Recruitment was primarily through a media campaign (86%) as well as speci c invitations to women who had responded to a mobile phone app focused on PND, originally developed in the USA 20 , and also ascertainment through the Pharmaceutical Bene ts Scheme prescription records for antidepressants, which requested participation from anyone with a depression diagnosis from a health professional. For further details of the recruitment strategy, see Byrne, et al. 19 .
AGDS participants were invited to complete an online questionnaire. A compulsory core module assessed self-reported psychiatric history, the Composite Interview Diagnostic Interview Short Form (CIDI-SF) which asseses the Diagnostic and Statistical Manual of Mental Disorders Fifth Edition (DSM-V) criteria for MDD 6 , and experiences of using commonly prescribed antidepressants. Women reporting symptoms of depression during pregnancy or up to 6 months following childbirth were asked to complete the lifetime Edinburgh Postnatal Depression Scale (EPDS) 21 , an adaptation of the standard EPDS 22 that assesses lifetime PND episodes. They were also asked whether symptoms of depression occurred during pregnancy, after giving birth, or both, the age at which they experienced their worst episode of PND, its severity and duration. For all AGDS participants, further voluntary modules assessed history of psychiatric health conditions and stressful life events. The AGDS protocol was approved by the Human Research Ethics Committee of QIMR Berghofer Institute for Medical Research.
Participants: PND cases and comparison groups Participants with major depression either met DSM-V criteria for MDD, or had been previously diagnosed with depression by a health professional. PND cases were de ned as women reporting at least one live birth who had been previously diagnosed with PND by a health professional, or who scored >= 13 on the lifetime EPDS, or who met criteria for major depression and reported at least one perinatal episode.
We identi ed two groups of cases, based on whether they had a history of MDD prior to their rst PND episode. Parous participants who reported an episode of depression before their rst pregnancy and met PND criteria (PND_priorDep) were compared with parous participants who reported an episode of depression before their rst pregnancy, but did not experience any depression associated with childbirth (NPD_priorDep). The second group comprised participants whose rst episode of depression occurred during the perinatal period (PND_ rstDep), compared to participants with depression onset at other times, but never during any peripartum period (NPD_all). The NPD_priorDep that forms the comparison group for the PND_priorDep sample is a subset of the NPD_all comparison group. It was expected that, given the early onset of major depression (before rst pregnancy) the NPD_priorDep would have more severe depression than the full NPD_all comparison group and might more closely match the PND_priorDep cases. Fig. 1   The outcome of interest was a PND episode for women with either a history of previous depressive episode(s), or no previous depression history. An exposure to a PND episode is de ned as the period of time from conception up to six months postpartum, so that the number of reported live births represents the number of exposures. The cross-sectional nature of our study meant that no direction of causality could be assessed, but we investigated risk factors for PND using variables that have previously been associated with PND 1,23,24 , including severity of depression; ancestry; comorbidity with other psychiatric disorders; adverse childhood experiences; reproductive traits and response to antidepressants.
We investigated previous history of depression as a modi er of the effect of each variable, by conducting two separate analyses of PND cases categorized as PND_priorDep or PND_ rstDep, for both descriptive and comparative measures. For comparative measures each of the two PND groups was compared to an appropriate comparison group. A further effect modi er is the time of onset of PND: during pregnancy, after delivery, or both. For both samples, a sensitivity analysis was conducted to investigate the effect of PND onset both during and after pregrancy.

Descriptive measures for cases
Clinical characteristics of PND cases included the length and severity of the worst PND episode. For both groups, the length of the PND worst episode was calculated, according to detailed occurrence before or after delivery. Length of the worst PND episode was measured using a 5 point scale: "Up to two weeks", "2-4 weeks", "1-3 months", "3-6 months", "More than 6 months". Details of occurrence included trimester of pregnancy or length of time after delivery.
Severity of the worst PND episode was measured using the level of interference with functioning, de ned as the need for any of the following: professional help, medication, and hospitalisation. More than one of the three measures could be chosen.

Comparative measures
Case and comparison groups were compared on a range of variables that have previously been identi ed to be associated with PND 1,23 . Demographic measures included current age, marital status, education and ancestry. A list of geographical regions from which ancestry is identi ed is provided in Table S2. More details are provided in Supplementary Methods. Clinical measures included the number and severity of episodes of major depression, history of childhood trauma and sexual or other physical assault, and previous diagnoses of psychiatric disorders. Eighteen psychiatric disorders were listed, of which twelve, identi ed by more than 3% of participants, were used in this study (Table S3). Reproductive measures included age at menarche, parity, age at rst birth, presence and severity of nausea and vomiting during pregnancy (NVP), and previous diagnosis of gestational diabetes, endometriosis or polycystic ovarian syndrome. Antidepressant measures included the number of antidepressants that had been tried, their e cacy and any side-effects. More details of these measures are provided in Supplementary Methods, which also lists the questions used to assess each characteristic.

Potential sources of bias
Two variables, number of births and age, signi cantly associated with PND, were identi ed as exposure and confounder respectively. Each birth represents an additional exposure to PND, whilst the negative association of PND with increasing age may re ect increasing awareness and diagnosis of the disorder, or imprecise memory of past events. Reliance on self-report information years after experiencing PND could lead to recall bias, although the inclusion of age as a covariate in regression analyses may alleviate this trend and participants who provided contradictory evidence were excluded from analysis. The lifetime EPDS used to assess PND is a screening, rather than diagnostic tool and may result in overestimation of PND case status 25 , although O'Connor et.al. 26 reported a sensitivity of 0.8 for the EPDS in identifying MDD with a cut-off score>=13, and a speci city of 0.9. The lifetime EPDS is a modi cation of this scale which has demonstrated internal consistency 21 .

Statistical Analysis
For both priorDep and rstDep groups, length and severity of the worst reported episode of PND was calculated, and logistic regression measured the association of depression length with early onset of PND ( rst trimester of pregnancy or within 4 weeks of delivery). Associations between variables and PND were assessed using logistic regression, with PND the dependent variable, separately for both priorDep and rstDep groups, including age at survey time and number of births as covariates.
All modules apart from the rst were optional, and some categories applied only to a limited number of participants (for example, those who had used at least one antidepressant). For these reasons, the number of participants who completed each category or variable varied. For each variable, the number of respondents is reported. Within each category, analysis employed Bonferroni correction for multiple testing (N=number of tests within each category).
Finally, to evaluate whether effect sizes were in uenced by time of PND onset, we conducted a sensitivity analysis that included only women who reported experiencing PND both before and after delivery. We conducted this analysis separately for both PND_priorDep and PND_ rstDep samples.
All analyses were conducted using R (version 3.6.0). Figures were generated using ggplot2 27 and Gliffy software 28 .

Results
Lifetime prevalence of depression during the peripartum period Just over 97% of AGDS participants (n= 20,191) reported previous diagnosis of depression by a health professional, of whom 88% met DSM-V criteria for MDD. The remaining 12% either did not complete the CIDI-SF, or did not meet DSM criteria. Of these participants with major depression, 75% (n=15,198) were female with median age of 39. Among female participants, 7,182 (47%) reported at least one live birth, and, of these, 5,058 (70%) met criteria for PND.
Of the 7,182 parous women, 2,933 reported a history of major depression prior to rst pregnancy. At least one episode of PND (PND_priorDep) was reported by 2,261 (77%) of these 2,933 women, whilst the remaining 672 women with no PND episodes (23%) formed their comparison group (NPD_priorDep). A total of 878 women reported that their rst episode of depression occurred during pregnancy or within the rst 6 months after delivery (PND_ rstDep), whilst all women who met criteria for major depression, had given birth to at least one child but did not satisfy criteria for PND (NPD_all, n=2,124) formed its comparison group. Of women who met criteria for PND, 1,919 were unable to be categorized as PND_priorDep or PND_ rstDep and were lost to further analysis. Fig. 1 and Supplementary Table S1 provide details of the sample selection process. Table 1 shows the reported time of onset of symptoms (for any PND episode) for both case groups (only during pregnancy, only after delivery, or both before and after delivery).

(2%)
The reported length of the worst episode of PND is shown in Fig.2 for PND_priorDep and in Fig.3 for PND_ rstDep. Full details are provided in Table S4. For both groups of cases, PND was most commonly reported to have lasted for more than six months. The most commonly reported time of PND onset for women whose episode began during pregnancy was during the rst trimester, and for those whose episode began after delivery was within 0-4 weeks. Both PND_priorDep and PND_ rstDep were more likely to report that their worst episode began after delivery (66% and 79% respectively), including 60% of PND_priorDep cases and 72% of PND_ rstDep cases who had reported that they experienced PND both before and after delivery. This difference between the groups is signi cant, with PND_ rstDep having 2.0 times the odds of reporting postpartum onset of worst case symptoms (CI=[1.7-2.4], P=4.6e-13) compared to the odds of PND_priorDep. For both groups, symptom onset in the rst trimester or 0-4 weeks postpartum was associated with longer duration of symptoms, signi cantly so for PND_priorDep (Table S4). For both groups, more than 60% required some sort of professional help, although less than 45% of women reported using medication to deal with this worst episode (Fig.4, Table S4).  were signi cantly associated with PND. Both age and number of births were included as covariates in subsequent analyses, which were also adjusted for multiple testing. Fig. 5 illustrates nominally signi cant results after the inclusion of covariates, with details of all results provided in Table S6.  reported more severe depression than NPD_priorDep (Fig. 5), although as expected, the NPD_priorDep comparison group also experienced signi cantly more severe depression than the NPD_all comparison group on all measures (Table S7).
Five of twelve psychiatric disorders (premenstrual dysphoric disorder (PMDD), attention de cit hyperactive disorder (ADHD), anxiety disorder, post-traumatic stress disorder (PTSD), and social anxiety disorder) were signi cantly associated with PND, although none survived Bonferroni correction. There was a signi cant association between PND and a history of self-reported childhood emotional abuse (OR=1. to report at least one side effect for antidepressants, compared with women with NPD_priorDep (including age, number of births and the number of antidepressants tried as covariates in the model) (Fig.  4). All of the 23 side effects were more commonly reported by PND_priorDep, 15 of them signi cantly so, although only 2 survived Bonferroni correction.
Clinical and psychosocial risk factors associated with PND as rst episode of depression.
As there may be unique risk factors associated with onset of depression perinatally, we conducted further analyses to evaluate differences between women who report their rst episode occurring perinatally (PND_ rstDep) and all NPD cases. Similar to priorDep ndings, we found that age at enrolment and number of births were associated with increased risk of PND ( Fig S2 and Table S6). After both these variables were included as covariates, PND_ rstDep was associated with emotional abuse during childhood, increased likelihood of trying at least 3 antidepressants compared with controls, and increased odds of reporting 13 of the 23 side effects, 5 of which were signi cant, although no side effects survived Bonferroni correction. No associations were found with other variables. FirstDep results (for variables that were nominally signi cant for priorDep) are illustrated in Fig.5 and full details of all results are provided in Supplementary Table S6.
Effect of PND onset on clinical and psychosocial risk factors associated with PND.
Symptoms of PND were experienced both during pregnancy and after delivery by 67% of PND_priorDep and 58% of PND_ rstDep. A sensitivity analysis using only these cases found that the odds ratios of variables already signi cantly associated with these groups increased. For priorDep, association of PND with three comorbidities: anxiety disorder, PTSD and social anxiety disorder remained signi cant after Bonferoni correction. Sexual abuse at any time became signi cantly associated with PND for priorDep as well as comorbidity with bipolar disorder, and PMDD became signi cantly associated with PND_ rstDep , although none of these survived Bonferroni correction. Full details are provided in Supplementary Table S8.

Discussion
We investigated lifetime prevalence and correlates of perinatal depression in a large cross-sectional study of depression. This is to date one of the largest studies of perinatal depression among women with major depression. Although previous research highlighted heterogeneity of PND 7,10 , until now detailed characteristics of women with PND but different psychiatric history have been lacking. Our study has enabled the identi cation of such characteristics through a comparison of two subsets of PND cases, with and without a prior history of major depression.
We found high lifetime prevalence of meeting criteria for probable PND in this sample, with the majority of women reporting symptoms both during and after pregnancy. Among those with prior history of major depression, PND was associated with more chronic, complicated depression, characterized by earlier onset, more reported episodes, more symptoms during the worst episode and increased likelihood of having a comorbid psychiatric disorder. They had signi cantly higher rates of reported emotional abuse and neglect and physical neglect during childhood, were more likely to report severe symptoms of NVP and suffer from more side effects to antidepressants. Women with no such prior history, whose rst depressive episode occurred during the perinatal period, did not report more severe depression, were no more likely to be comorbid with other psychiatric disorders, apart from PMDD, and no more likely to report severe NVP than women who experienced depression outside the perinatal period. Like PND cases with a prior history of depression, women who experienced PND as their rst depressive episode reported signi cantly more side effects to antidepressants than women with depression without a perinatal episode, and were also more likely to report childhood emotional abuse.
The main limitation of this study is that it is based on an online questionnaire, with no personalized interviews or clinical reports to provide supporting evidence for self-reported data. Answers were based on total life experience, including, but not exclusive to, the perinatal period. Furthermore, the AGDS is a cross-sectional study. Its strength lies in its sample size, but, unlike a longitudinal study, it provides no information with respect to timing of variables signi cantly associated with PND (with the exception of childhood adverse experiences), so no inference can be made pertaining to cause and effect. The results of the study should also be considered in the context that the AGDS cohort is mostly young and welleducated, and may not generalize to the entire population.
Despite these limitations, the ndings of this study are consistent with previous reports. PND for women with a previous history of depression seems to be more severe and complex than for women who experience PND as rst depression onset, supporting the notion of PND heterogeneity according to previous psychiatric history. Prior history of psychiatric disorders, stress, and a history of abuse have emerged as strong predictive factors for PND 1,[7][8][9][10]12,29 . PMDD is the severe form of premenstrual syndrome, recently identi ed as a risk factor for PND 30 , and NVP has been recognized as the strongest obstetric predictor of PND 31 . Previous studies have also found that women suffering from both MDD and PND had more severe depression and higher incidence of anxiety disorder and childhood trauma than women suffering from MDD alone 21 , and that most severe depression is suffered by women who experience PND both during pregnancy and after delivery 10 .
This study found high reported rates of non-response to antidepressants in women experiencing PND for both subgroups. Studies of the e cacy of antidepressants for the treatment of PND have been inconclusive 32 , and, to our knowledge, increased incidence of side effects amongst women with PND has not been previously reported. Further clinical studies of antidepressant e cacy in PND are warranted, as well as e cacy of alternative treatments.
Finally, complications of pregnancy and birth were not assessed in this study apart from NVP and gestational diabetes, so it was not possible to fully assess whether perinatal complications may contribute to PND vulnerability 7 .

Conclusions
PND is a leading cause of disease for women who give birth, adding to the overall family disease burden and potential cognitive and emotional problems for affected children. This sample of parous women with lifetime major depression found a high rate of PND, particularly for women who experienced an episode of depression before their rst pregnancy. There is a compelling literature demonstrating that screening for PND should begin during pregnancy 26,33 , particularly for women with prior history of depression, which is supported by the nding that the majority of cases in this study experienced PND both before and after delivery. Although women who have been previously diagnosed with major depression are, presumably, under clinical care, it is possible that women may have withdrawn from care if treatment has been ineffective. Standard prenatal care that adopts frequent assessment of depression status provides an opportunity to identify women who might otherwise "slip through the cracks" and ensure that they continue to receive support in nding a successful treatment or in the prevention of relapse 26 . Our results also support the screening of childhood adverse experiences and PMDD in pregnancy, given that all women with PND in this study had increased odds of a history of emotional abuse and neglect, as well as increased odds of PMDD. Cases were also more likely to have treatment resistant depression, with increased odds of side effects, supporting further clinical investigation of antidepressant e cacy in PND.

Competing interests
No con ict of interest has been reported Availability of data and material Data used in this analysis and described in this article are available to all interested researchers through collaboration. Please contact NGM.
Ethics approval and consent to participate All study protocols were approved by the QIMR Berghofer Medical Research Institute Human Research Ethics Committee. The protocol for approaching participants through the DHS, enrolling them in the study, and consenting for all phases of the study (including invitation to future related studies) and accessing MBS and PBS records was approved by the Ethics Department of the Department of Human Services.
Patient consent for participation in the study was obtained.  Flow chart: Selection of cases and associated comparative group for rst analysis (prior history of major depression) and second analysis (PND is rst experience of major depression). Cases met criteria for major depression and had at least one live birth, plus any of: EPDS score >=13; a previous diagnosis of PND; or major depression during the perinatal period. Of the comparison groups, NPD_priorDep is a subset of NPD_all. NPD_priorDep was considered to be a more appropriate comparison group for PND_priorDep since members of both these groups experienced an episode of major depression before rst pregnancy.  Forest plot of odds ratios with con dence intervals of all variables nominally signi cantly associated with PND_priorDep cases when compared with NPD_priorDep. Odds ratios for the association of these variables with PND_ rstDep cases compared with NPD_all are also included for comparison. Logistic regression, including age and number of live births of participants as covariates, was used to calculate odds ratios.