Prediction of postpartum depression by sociodemographic, obstetric and psychological factors: A prospective study
*Yong-Ku Kim, MD, PhD, Department of Psychiatry, Korea University College of Medicine, Ansan Hospital, Ansan City, Gojan Dong, 516, Kyunggi Province 425-020, Korea. Email: email@example.com
Aim: Many studies have documented serious effects of postpartum depression. This prospective study sought to determine predictive factors for postpartum depression.
Methods: Pregnant women (n = 239) were enrolled before 24 weeks in their pregnancy. At 6 weeks postpartum, 30 women who had postpartum depression and 30 non-depressed mothers were selected. The Edinburgh Postnatal Depression Scale (EPDS), the Beck Depression Inventory (BDI), the Beck Anxiety Inventory (BAI), the Rosenberg Self-Esteem Scale (RSES) Marital Satisfaction Scale (MSS), and the Childcare Stress Inventory (CSI) were administered to all 60 mothers at 24 weeks pregnancy, 1 week postpartum, and 6 weeks postpartum.
Results: The differences in most of the diverse sociodemographic and obstetric factors assessed were not statistically significant. There were significant differences in MSS scores at 24 weeks pregnancy (P = 0.003), and EPDS (P < 0.001; P = 0.002), BDI (P = 0.001; P = 0.031), and BAI (P < 0.001; P < 0.001) at both 24 weeks pregnant and 1 week postpartum, while there was no significant difference in the RSES scores at 24 weeks pregnant (P = 0.065). A logistic regression analysis was performed on the following factors: ‘depressive symptoms immediately after delivery’ (EPDS and BDI at 1 week postpartum), ‘anxiety’ (BAI prepartum), ‘stress factors from relationships’ (MSS prepartum and CSI at 1 week postpartum) or ‘self-esteem’ (RSES prepartum). When these four factors were added individually to a model of the prepartum depressive symptoms (EPDS and BDI prepartum), no additional effect was found.
Conclusions: The optimum psychological predictor is prepartum depression, and other psychological measures appear to bring no significant additional predictive power.
POSTPARTUM DEPRESSION (PPD) is diagnosed when a mother shows the onset of mood episodes within 4 weeks postpartum.1 Postpartum depression occurs in 10–20% of delivered mothers; childbirth creates a stressful period that may interact with innate vulnerability and environmental factors.2,3 Because of its adverse effects upon the cognitive, social, and emotional development of the child, PPD is considered to be a critical public health problem.
Although many studies have tried to explore the psychosocial predictors of PPD, and have pointed to prepartum depression as a major factor,4 some researchers have focused on sociodemographic factors such as financial/professional difficulties,5 early loss of a parent,6,7 stressful events during pregnancy and delivery,8 and obstetric difficulties.9 Other researchers have emphasized psychological signs such as anxiety,10 marital dissatisfaction,11 self-esteem issues,12 and psychological conditions.13
However, there have been few studies that have controlled for variations in prepartum depressive symptoms which have shown a large effect size (0.49 to 0.51).14 Moreover, most studies have been limited by the use of retrospective methods, small sample sizes, or diagnosis based on self-report scales, making it difficult to comprehend the effects of prepartum conditions and predictive factors.15,16
In this study, statistical differences in sociodemographic and psychological features were explored with the intent of determining whether any other factors besides prepartum depressive symptoms have significant implications on diagnosis. We recruited a large number of pregnant women before their deliveries for prospective observation. After that, we assessed the mothers' sociodemographic and obstetric data during pregnancy, and investigated psychological factors three times between 24 weeks pregnancy to 6 weeks postpartum, because repeated measures would suggest more reliable predictive factors.17 Finally, we administered logistic regression analyses to examine the additional effects of the above factors on the prediction of PPD.
We recruited pregnant women before 24 weeks of pregnancy through an obstetrician in the Department of Obstetrics and Gynecology of the Sungkyunkwan University Medical Center in Seoul, South Korea. A total of 239 women gave informed consent and were enrolled.
Following the diagnostic criteria on DSM-IV, ‘onset of episode within 4 weeks postpartum’, the diagnosis of postpartum depression was made at 6 weeks after delivery. Thirty mothers were classified as having postpartum depression based on the criterion of having a score ≥9.5 on the Korean version of Edinburgh Postnatal Depression Scale (K-EPDS).18 The K-EPDS was administered at 24 weeks pregnancy, 1 week after delivery, and 6 weeks after delivery. This score (≥9.5) is the optimal cut-off in South Korea. The depressed mothers (K-EPDS ≥ 9.5) were given a semi-structured interview using Structured Clinical Interview for DSM-IV (SCID-I) by a trained psychiatrist. The SCID-I confirmed that all 30 mothers had clinical depression; thus, the postpartum depression group was confirmed as 30 mothers.
To select a group of non-depressed mothers for comparison, we selected 30 healthy mothers who had the lowest K-EPDS scores (<3) in ascending series from among the candidates. The K-EPDS was administered at 24 weeks pregnancy, 1 week after delivery, and 6 weeks after delivery. Exclusion criteria included any severe or unstable medical illness.
This study was reviewed and approved by the Ethics Committee of the Korea University Medical Center.
The psychiatric status of each of the 30 patients with depression was assessed by the SCID-I. For all subjects, information on sociodemographic data were investigated by means of structured questionnaires. Each interview lasted about 45 min or less.
Questionnaires about sociodemographic, obstetric, and psychiatric features were administered at 24 weeks pregnancy and at 1 week postpartum (Tables 1–3). The questions asked prepartum were related to age, marriage, education, occupation, income, housing, and psychiatric history. The questions asked postpartum were related to the experience of pregnancy and delivery, the season of delivery, difficulty of the pregnancy, the desired sex of the baby, the actual sex of the baby, psychological conflict among the family, ownership of the house, experiences of parental loss, medical illness, history of depression, psychiatric treatment, method of delivery, breast or bottle feeding, family history, companionship in the delivery room, place of postpartum care, living arrangements, help with care of the baby, the health of the baby, and the baby's crying.
Table 1. Sociodemographic characteristics (documented at 24 weeks pregnancy)
| 30–39||20||66.7||24||80.0|| || |
| Unmarried||2||6.7||1||3.3|| || |
| 7–9||5||16.7||6||20.0|| || |
| 10–12||21||70.0||20||66.7|| || |
| ≥13||2||6.7||1||3.3|| || |
| No Job||5||16.7||6||20.0||0.101||0.904|
| Part-time||2||6.7||3||10.0|| || |
| Full-time||23||76.7||21||70.0|| || |
| No Job||14||46.7||18||60.0||2.166||0.128|
| Part-time||7||23.3||1||3.3|| || |
| Full-time||9||30.0||11||36.7|| || |
| 1–1.9 million||7||23.3||6||20.0||4.788||0.006*|
| 2–2.9 million||2||6.7||13||43.3|| || |
| 3–3.9 million||16||53.3||3||10.0|| || |
| ≥4 million||5||16.7||8||26.7|| || |
| Husband's parents' house||9||30.0||5||16.7||0.689||0.508|
| Woman's parents' house||0||0||1||3.3|| || |
| Married couple and children only||21||70.0||24||80.0|| || |
Table 2. Obstetric features (documented at 1 week postpartum)
| One||12||40.0||9||30.0|| || |
| Two or more||0||0.0||0||0.0|| || |
|Difficult of pregnancy|
| In vitro fertilization||1||3.3||0||0.0||0.817||0.493|
| Fertility drug||3||10.0||0||0.0|| || |
| Folk remedy||1||3.3||0||0.0|| || |
| None||25||83.3||30||100.0|| || |
| Mild||21||70.0||8||26.7|| || |
| Moderate||7||23.3||8||26.7|| || |
| Severe||2||6.7||5||16.7|| || |
|Season of delivery|
| Summer||6||20.0||3||10.0|| || |
| Autumn||3||10.0||7||23.3|| || |
| Winter||15||50.0||10||33.3|| || |
| With spouse||10||33.3||10||33.3|| || |
| With other family member||4||13.3||3||10.0|| || |
| Natural childbirth||18||60.0||22||73.3||0.961||0.332|
| Caesarean section||12||40.0||8||26.7|| || |
| No||18||60.0||10||33.3|| || |
| Girl||14||46.7||14||46.7|| || |
|Coincidence of baby's gender and gender hoped for|
| No||4||13.3||5||16.7|| || |
| Bottle feeding||6||20.0||2||6.7|| || |
| Mixed feeding||5||16.7||6||20.0|| || |
| Husband's parents' house||13||43.3||7||23.3||1.217||0.314|
| Woman's parents' house||2||6.7||2||6.7|| || |
| One's own house||8||26.7||15||50.0|| || |
| Postpartum care agent||7||23.3||6||20.0|| || |
|Current dwelling place|
| Husband's parents' house||8||26.7||6||20.0||0.371||0.774|
| Woman's parents' house||2||6.7||1||3.3|| || |
| One's own house||16||53.3||20||66.7|| || |
| Other||4||13.3||3||10.0|| || |
| By oneself||9||30.0||2||6.7||2.447||0.076|
| With husband's parents||1||3.3||7||23.3|| || |
| With woman's parents||16||53.3||15||50.0|| || |
| Babysitter||4||13.3||6||20.0|| || |
|One's own health|
| Better than before||1||3.3||2||6.7||1.594||0.203|
| No change||12||40.0||20||66.7|| || |
| Worse than before||16||53.3||1||23.3|| || |
| Worst than ever||1||3.3||7||3.3|| || |
| Far better than expected||6||20.0||7||23.3||0.226||0.878|
| Rather better than expected||16||53.3||14||46.7|| || |
| Same as expected||4||13.3||6||20.0|| || |
| Rather worse than expected||4||13.3||3||10.0|| || |
| Far worse than expected||0||0.0||0||0.0|| || |
Table 3. Psychiatric features and related variances (documented at 24 weeks pregnancy)
|History of depression|
| No||28||93.3||28||93.3|| || |
|History of psychiatric medication|
| No||20||66.7||28||93.3|| || |
| No||21||70.0||22||73.3|| || |
|Family history of mental disorder|
| Yes||0||0.0||0||0.0|| || |
| No||30||100.0||30||100.0|| || |
|Experience of parental loss|
| No||25||83.3||28||93.3|| || |
|Mental conflict between family|
| Husband's family||13||43.3||6||20.0||0.703||0.505|
| Husband||4||13.3||4||13.3|| || |
| Others||13||43.3||20||67.7|| || |
|Source of prepartum anxiety|
| Fear of delivery||14||46.7||3||10.0||0.994||0.423|
| Fear of indisposition||5||16.7||7||23.3|| || |
| Anxiety about childrearing||9||30.0||19||63.3|| || |
| Loss of attractiveness||2||6.7||1||3.3|| || |
| Not applicable||0||0.0||0||0.0|| || |
The Edinburgh Postnatal Depression Scale (EPDS) is a self-report scale designed to detect postnatal depression.19 The EPDS was translated into Korean and standardized.18 The questions ask about mood (eight items) and anxiety (two items), and each item is rated on a 4-point scale. All mothers were given the EPDS at 24 weeks pregnancy, 1 week postpartum, and 6 weeks postpartum.
The Beck Depression Inventory – Korean Version (BDI) was developed to assess not only depressive symptoms and their severity, but also motivational and behavioral symptoms.20 Lee and Song standardized the BDI to include 21 items in the Korean version.21 The BDI was administered at 24 weeks pregnancy, 1 week postpartum, and 6 weeks postpartum (α = 0.78).
The Beck Anxiety Inventory (BAI)22 subjectively measures the level of anxiety that an individual experiences. The self-report scale includes 21 items concerning anxiety experienced during the past 1 week. The scale, as translated into Korean by Kwon,23 was used at 24 weeks pregnancy, 1 week postpartum, and 6 weeks postpartum (α = 0.94).
The Childcare Stress Inventory (CSI)24 assesses stressful postpartum events related to childcare. The items include health and feeding problems with the baby and satisfaction with social support provided by the spouse and family. We used the Korean version of the scale developed by Chun2 at 1 week postpartum and 6 weeks postpartum (α = 0.90).
The Rosenberg Self-Esteem Scale (RSES)25 assesses statements that represent self-esteem. The Korean version of the scale adopted by Lee26 asks mothers to respond to 11 items, such as ‘I am generally satisfied with myself’, on a 5-point Likert scale. This assessment was given at 24 weeks pregnancy (α = 0.77).
The Marital Satisfaction Scale (MSS)27 was developed to evaluate an individual's satisfaction with their marriage and spouse. Lee constructed a 48-item scale in Korean and administered the validity study.28 Marital satisfaction, an attitude of favorability toward one's own marital relationship, was assessed at 24 weeks pregnancy along with the RSES (α = 0.951).
Data were analyzed using Student's t-test or Fisher's χ2 test according to the characteristics of the variables. Pearson's correlation coefficients between psychiatric scales were tested. Discriminate power on postpartum depression was tested with logistic regression analyses. All data were processed by SPSS 12.0 for Windows (SPSS Inc, Chicago, IL, USA).
There were statistically significant differences between groups in the EPDS (P < 0.001), BDI (P < 0.001), and BAI (P < 0.001) at 6 weeks postpartum, the diagnosis point (Table 4). All sociodemographic, obstetric and psychiatric variables and differences between the postpartum depression group and the non-depressed group are shown in Tables 1–3. The differences in age, marriage, education level, former occupation, present occupation, experience of pregnancy or delivery, season of delivery, difficulty of pregnancy, desired sex of the baby, actual sex of the baby, psychological conflict among the family, ownership of the house, experiences of parental loss, medical illness, history of depression, psychiatric treatment, method of delivery, breast or bottle feeding, family history, companionship in the delivery room, place of postpartum care, living arrangements, help with baby care, health of the baby, and the baby's crying were not statistically significant. However, there were significant differences in income (P = 0.006) and nausea (P < 0.001).
Table 4. Differences in clinical scales between women with and without postpartum depression
Clinical features between depressed and non-depressed postpartum mothers
There were significant differences between the two groups in the scores on the MSS (P = 0.003 at pregnancy 24 weeks), the EPDS (P < 0.001 at pregnancy 24 weeks; P = 0.002 at 1 week postpartum), the BDI (P = 0.001 at pregnancy 24 weeks; P = 0.031 at 1 week postpartum), and the BAI (P < 0.001 at pregnancy 24 weeks; P < 0.001 at 1 week postpartum) (Table 4). There was no significant difference in the RSES scores at 24 weeks pregnancy (P = 0.065) (Table 4).
Correlations between psychiatric scales
Pearson's correlation coefficients between psychiatric scales were calculated (Table 5). Almost all scales had significant correlations with each other. However, we found no significant correlations between the EPDS at 1 week postpartum and the MSS score (P =−0.275), between the RSES and the two EPDS scores (P =−0.147 at 24 weeks pregnancy; P = −0.190 at 1 week postpartum, respectively), between the EPDS at 24 weeks pregnancy and the three BDI scores (P = 0.139 at 24 weeks pregnancy; P = 0.195 at 1 week postpartum; P = 0.108 at 6 weeks postpartum, respectively) or CSI at 1 week postpartum (P = 0.303) or between the EPDS at 1 week postpartum and the BDI/BAI at 24 weeks pregnancy (P = 0.207; P = 0.259, respectively).
Table 5. Correlations between clinical scales
|EPDS w1||0.337*|| || || || || || || || || || || |
|EPDS w6||0.404**||0.544**|| || || || || || || || || || |
|BDI w0||0.139||0.207||0.635**|| || || || || || || || || |
|BDI w1||0.195||0.347*||0.513**||0.537**|| || || || || || || || |
|BDI w6||0.108||0.316*||0.670**||0.611**||0.577**|| || || || || || || |
|BAI w0||0.569**||0.259||0.603**||0.497**||0.514**||0.386*|| || || || || || |
|BAI w1||0.333*||0.584**||0.707**||0.730**||0.595**||0.397**||0.734**|| || || || || |
|BAI w6||0.433**||0.458**||0.763**||0.564**||0.465**||0.542**||0.691**||0.828**|| || || || |
|CSI w1||0.303||0.306*||0.613**||0.579**||0.488**||0.537**||0.508**||0.640**||0.721**|| || || |
|CSI w6||0.471**||0.197||0.762**||0.495**||0.455**||0.643**||0.560**||0.615**||0.801**||0.693**|| || |
|RSES w0||−0.147||−0.190||−0.376*||−0.758**||−0.556**||−0.246||−0.381*||−0.544**||−0.353*||−0.367*||−0.233|| |
Predictors of postpartum depression
A logistic regression analysis was used to explore the psychological features that could help predict postpartum depression. We classified the factors into four types: ‘depressive symptoms immediately postpartum’ (EPDS and BDI at 1 week postpartum), ‘anxiety’ (BAI at 24 weeks pregnancy and 1 week postpartum), ‘stress factors from relationships’ (MSS prepartum and CSI at 1 week postpartum), and ‘self-esteem’ (RSES), with the intent to exclude the effects of covariance between psychological factors. After that we entered each factor into a model.
Model incorporating depressive symptoms immediately postpartum
In the first regression analysis of predictors of postpartum depression, the EPDS and the prepartum BDI scores for depression were entered into the first model (Model 1), and this was considered to be the baseline (Table 6). In the next step, the EPDS and the BDI at 1 week postpartum were entered into a second model to determine the effect of depressive symptoms immediately after delivery on postpartum depression, in addition to depression during pregnancy. Based on this, the EPDS and the BDI before delivery correctly identified 96.7% of depressed mothers and 76.9% of non-depressed mothers (Model 1; B = 0.77, Wald = 5.45, P < 0.05 and B = 0.35, Wald = 0.16, P < 0.05, respectively). However, when the additional variables, the EPDS and the BDI at 1 week postpartum, were added to model 1, they did not show significant differences (Model 2; B = −0.01, Wald = 0.00, P = 0.97 and B = 0.27, Wald = 0.60, P = 0.44, respectively).
Table 6. Logistic analyses for predicting postpartum depression
|Model 1. Prepartum depression|
| EPDS w0||0.77||0.33||5.45||0.020||2.16 (1.13–4.12)|
| BDI w0||0.35||0.16||4.96||−0.026||1.42 (1.04–1.93)|
|Model 2. Postpartum depression|
| EPDS w1||−0.01||0.16||0.00||0.97||0.99 (0.72–1.37)|
| BDI w1||0.27||0.35||0.60||0.44||1.31 (0.66–2.57)|
|Model 1. Prepartum depression|
|Model A. Anxiety|
| BAI w0||0.21||0.12||3.23||0.07||1.23 (0.98–1.55)|
|Model 1. Prepartum depression|
|Model B. Stress form relationship|
| MSS||1.58||102.23||0.00||0.99||4.83 (0.00–5.027E + 87)|
| CSI||6.32||333.85||0.00||0.99||552.97 (0.00–8.198E + 286)|
|Model 1. Prepartum depression|
|Model C. Self-esteem|
| RISP||0.09||0.15||0.37||0.54||1.10 (0.82–1.47)|
Model incorporating anxiety
The BAI score was added to Model 1 to test the effect of anxiety, and this was named Model A. The state of anxiety had no additive effect compared with Model 1, which considered only prepartum depressive symptoms (B = 0.21, Wald = 3.23, P = 0.07; Table 6).
Model incorporating stress from relationships
The MSS and the CSI scores, which represent the factors of relationship stress, were added to Model 1 to produce Model B (Table 6). The comparison of Model 1 and Model B demonstrated that MSS and CSI had no significant additional effects on the prediction of postpartum depression (B = 1.58, Wald = 0.00, P = 0.99 and B = 6.32, Wald = 0.00, P = 0.99, respectively).
Model incorporating self-esteem
The variable of self-esteem, represented by the RSES score, was added to produce Model C. As in the preceding results, Model C, which considered both prepartum depression and self-esteem as predictors, was not significantly more effective than Model 1 (B = 0.09, Wald = 0.37, P = 0.54; Table 6).
At the beginning of the present study, 239 pregnant women were enrolled. To detect specific predictive signs or risk factors, sociodemographic data and obstetric information were investigated. In addition, the psychological characteristics were identified prospectively at 24 weeks pregnancy, 1 week postpartum, and 6 weeks postpartum. At 6 weeks postpartum, 30 of the 239 mothers were diagnosed as having postpartum depression based on SCID-I scores. We selected 30 non-depressed mothers of the same subject pool as a comparison group.
The study findings indicate that prepartum depression has good predictive power on postpartum depression. However, there were no additional effects when sociodemographic or obstetric data and other psychological features were considered in the prediction of postpartum depression.
In the study, we failed to find sociodemographic or obstetric predictors that could provide a significant discrimination of the postpartum depression group from the non-depressed group. Although significant differences were found in incomes and nausea (P = 0.006, P < 0.001, respectively), these variances could not be taken as risk factors in that there was no linear feature or distinctive aspect to interpret (Table 1). These results support prior studies that found no relationship between postpartum depression and demographic factors such as age, education, or income.17,29,30
The EPDS, BDI, and BAI scores during pregnancy and immediately after delivery, as well as the MSS during pregnancy and the CSI at 1 week postpartum, showed statistical differences between each other (Table 5). Even though there was no significant difference in the RSES scores between groups, it is not surprising when considering that a previous study suggested that there is no significant relationship between a mother's depression and self-esteem.31 However, the effect of self-esteem on a mother's depression should be examined in further studies, because other researchers have asserted that self-esteem does play a significant role in the onset of clinical depression.32
In the analysis of sensitivity, specificity, and predictive value, the BAI score that was measured at 24 weeks pregnancy had the highest discriminative power (classified 83.7%) among any of the scales, including the BAI at 1 week postpartum (78.0%). Also, the EDPS score at 24 weeks pregnancy accurately classified 79.1% of the depressed mothers, and the EPDS at 1 week postpartum classified 72.5%. The BDI at 24 weeks pregnancy classified 76.7%, and at 1 week postpartum classified 66.7%. The CSI at 1 week postpartum and the RSES at 24 weeks pregnancy classified 72.2% and 72.1% of PPD subjects, respectively. Using the MSS, 75.8% were accurately classified. However, a more important question is which psychological feature, when added to the data on prepartum depressive symptoms, would have a good predictive power for anticipating PPD in a clinical setting. Because depressive symptoms during pregnancy are better correlated to PPD than are any other features, even if another psychological factor had predictive power, we could not consider that factor in isolation to the exclusion of depressive symptoms.
For this reason, logistic regression, a powerful tool for analyzing patient characteristics, was performed to identify the most beneficial variables in predicting postpartum depression. Astonishingly, when ‘depressive symptoms immediately after delivery’ (EPDS and BDI at 1 week postpartum), ‘anxiety’ (BAI prepartum), ‘stress factors from relationships’ (MSS prepartum and CSI 1 week postpartum) or ‘self-esteem’ (RSES prepartum) were added to the prepartum depressive symptoms (EPDS and BDI prepartum), there was no additional effect in the prediction of postpartum depression. These results raise doubts about whether any incidental information besides prepartum depression should be collected and examined to predict postpartum depression, particularly when considering the effort and expense required gathering information from women around the time of childbirth. As shown in Table 5, and in prior studies, the links between depression, anxiety, and other psychological features are prominent, as shown by the increased covariance between variables.33 For these reasons, we would rather focus on prepartum depression, which can be measured by the EPDS and BDI tests, and save resources by not attempting to tap into many factors that have no additional advantage for prediction.
Our findings have clinical implications and consequences about the economical and efficient prediction of postpartum depression as follows. First, it is hard to find any distinctive sociodemographic or obstetric indicators, even considering the mother's anxiety, relationship stress, marital satisfaction, and self-esteem, other than prepartum depression. Therefore, we need to pay close attention to women's depressive symptoms during pregnancy more than to any other factor.34 Second, we used a prospective study, which is the most reliable method to discover the effect of predictive factors on postpartum depression; prior to this study, the prospective method has been poorly applied in Asia, including South Korea. In addition, we selected the groups of women with and without postpartum depression from a large pool in order to minimize the noise in the diagnosis of postpartum depression and make clear the characteristics of both groups. Third, in our prospective study, going one step beyond a simple comparison, logistic regression was applied. This study, which closely examined various factors on PPD to the exclusion of the effect of prepartum depression, could be a great help to determine the most efficient way to predict the disorder.
There are two limitations that should be taken into account in interpreting the findings in this study. First, 60 women comprise the final subjects of this study. Further study will be needed using more subjects and diverse scales in order to assure more robust results. Second, prepartum depression, which was found to be a good predictor, was measured only once, at 24 weeks pregnancy. Future research should also aim to assess various time points during pregnancy in order to find the assessment time frame with the best predictive value for postpartum depression.
Funding for this study was provided by a grant from the Psychiatric Research Foundation of Korean Neuropsychiatric Association (KNPA).