Postpartum depression: identification of women at risk


Correspondence: Dr D. N. Forman, Forteledet 16, st, DK 8240 Risskov, Denmark.


Objective To identify and test the predictive power of demographic, obstetric, and psychosocial risk factors of postpartum depression.

Design Community-based, prospective follow up study based on questionnaires on past history of psychiatric disease, psychological distress and social support during pregnancy and depression at four months after delivery. Obstetric files were collected at time of birth.

Setting Antenatal care clinic and delivery ward, Aarhus University Hospital, Denmark.

Population 6790 women giving birth between 1 January 1994 and 31 December 1995, who attended the antenatal clinic during pregnancy; 5252 (78%) completed all questionnaires. The validation population comprised 528 women enrolled immediately prior to and after the study period.

Main outcome measure Postpartum depression four months after giving birth assessed by the Edinburgh Postnatal Depression Scale.

Results 5.5% of the women suffered from postpartum depression, corresponding to a score of 13 or higher on the Edinburgh Postnatal Depression Scale. Risk factors identified by multivariate logistic regression analysis included psychological distress in late pregnancy (OR 6.3 [95% CI 4.4–9.1]), perceived social isolation during pregnancy (OR 3.6 [95% CI 1.9–7.0]); high parity (OR 3.8 [95% CI 1.8–8.0]); and a positive history of prepregnant psychiatric disease (OR 2.1 [95% CI 1.4–3.2]). No association was found between pregnancy or delivery complications, and postpartum depression. The maximum predictive power of the identified risk factors was 0.3. According to these results, one out of three women who suffers from psychological distress in late pregnancy with perceived social isolation will develop postpartum depression.

Conclusion Antenatal focus on psychosocial wellbeing may help to identify women at risk of postpartum depression.


Nonpsychotic depression in the postpartum period is a well known clinical phenomenon, but women at risk are rarely recognised during pregnancy or at the delivery ward1,2. Prevalence varies between 8% and 15% according to the different diagnostic criteria used2–7. Although the incidence in the year following childbirth does not appear to exceed the incidence of depression found in nonchildbearing populations8,9, a public health problem exists. Postpartum depression has a long term effect on mental health since it may increase the risk of continuing or recurrent depression3,10–13. Postpartum depression has also been associated with adverse effects on early infant development14–18, especially among socially disadvantaged children. Serious consequences for the child include increased risk of accidents, sudden infant death syndrome, and an overall higher frequency of hospital admissions19–21.

Numerous studies have addressed the significance of various biological and nonbiological factors in the aetiology of postpartum depression. The frequent use of retrospective study methods may explain the conflicting data available on the role of perinatal events22–27. The most consistent findings link low socioeconomic status and previous psychiatric illness to postpartum depression5,28–33. Few attempts have been made to test the predictive power of putative risk factors32,34, and none have used standardised and well validated questions. The aim of the present study was to test the value of potential risk factors measured during pregnancy and prenatal period in the prediction of postpartum depression.


The population comprised a cohort of Danish speaking women attending the antenatal programme for second trimester examination. The women were delivered at the Department of Obstetrics at Aarhus University Hospital in the period from December 1993 to March 1996. Development of predictive index was based on data collected from 1 January 1994 to 31 December 1995. The model was tested using data collected in the four months immediately prior to and after this period.

The study cohort comprised 6790 women who were enrolled about the 16th week of gestation. All women were asked to complete the first questionnaire which requested age, parity, socioeconomic status, medical, obstetric, and psychiatric history, use of alcohol and drugs. The questionnaire was completed at home. Two reminders were mailed to nonrespondents. At the 30th week of gestation, the women who had returned the first questionnaire were sent a second questionnaire which asked about social support, psychological distress and intake of alcohol. One reminder was mailed to nonrespondents.

Screening for postpartum depression was performed at four months postpartum by a third self-administered questionnaire mailed to women who had answered the two first questionnaires. Women who had experienced pre-, peri- or postnatal fetal death were excluded.

Information about pregnancy and delivery was obtained from a specific registration form filled in by the midwife immediately after delivery. All obstetric data were validated by a research midwife reviewing the hospital records.


The main outcome measure was the Edinburgh Postnatal Depression Scale, which has been validated for detection of postpartum depression in community samples35,36. Using a threshold score of 12/13, the scale has been found to have a sensitivity ranging from 68% to 86% and specificity from 78% to 96%35,37. The Edinburgh Postnatal Depression Scale was translated into Danish and then back-translated in order to check that the questions elicited the intended information.

The principal measure of psychological distress in the 30th week of gestation was the 12-item version of the General Health Questionnaire, a screening instrument designed for use in general population surveys38. This score indicates the severity of psychological distress on a continuous scale. Content and concurrent validity of the 30-item version has been tested in a Danish population39,40. Several studies have found the General Health Questionnaire an appropriate method of measuring psychological distress among pregnant women41,42.

The General Health Questionnaire score was categorised into three levels of distress according to Goldberg38. The principal measure of social support was based on the subjective feeling of perceived social isolation, categorised into three levels: never, occasionally and often. Social support was initially covered by 15 questions. The questions addressed number of social contacts, and perceived social support from close iends and family. However, analyses revealed the single question to be most informative and strongest associated to postpartum depression.

Analyses were conducted in three phases. Potential risk factors were identified in the study population of 6790 women. Subsequently, a predictive index was developed and tested on a smaller validation sample of 528 women enrolled in the months immediately before and after the study period.

In the first phase, potential variables were evaluated for bivariate association with postpartum depression, and significant associations were subsequently entered into multivariate logistic regression analyses. The predictive power of the risk factors was summarised as a risk index, defined as a sum of the risk factors weighted by the estimated regression coefficients obtained from the logistic regression analysis. A risk index was determined for each woman. Two predictive indices were considered: one index involving only antenatal risk factors, the other involving both antenatal and perinatal risk factors. The predictive indices were applied to the validation sample in order to test validity coefficients of sensitivity, specificity and positive predictive values. Receiver-operator characteristic curves were constructed43.

Statistical methods

Bivariate associations were tested with Student's t test for normally distributed continuous variables, and the Mann-Whitney U tests for continuous variables not normally distributed. The χ2 test was used to assess associations between postpartum depression and categorical risk factors. Logistic regression was used when multiple variables were considered simultaneouly. Odds ratios (OR) were presented with 95% confidence intervals (CI). Statistical significance was defined as two-sided P values using a significance level of 5%. The area under the receiver-operator curve was derived from the Mann-Whitney U statistic for comparing the distribution of the risk index among depressed and nondepressed women44.

The project was approved by the regional committee for ethics in science.


Of the 6790 women enrolled in the study cohort, 6388 (94%) completed the first questionnaire; 5866 (86%) completed the second questionnaire; 5262 women (78%) completed all three questionnaires. Twenty-eight women were not contacted because of child death, and nine women were lost to follow up.

The screening questionnaire for postpartum depression was completed by 5091 women who did not miss any items. Multivariate analysis was performed on fewer women because a number of women had missing data for one or more variables. Over the two year follow up period, 85 women were enrolled twice due to repeated pregnancies. In these cases only data from the first pregnancy were used. A score of 13 or above was achieved by 281 women, indicating a point-prevalence of postpartum depression of 5.5 (281/5091).

Distribution of study variables among depressed and nondepresssed is shown in Tables 1–3. The variables are grouped into sociodemographic characteristics, psychopathological and social support status, and perinatal events. Bivariate associations are provided as unadjusted odds ratios for all study variables.

Table 1.  Sociodemographic characteristics by postpartum outcome. Values are given as %, unless otherwise stated.
VariableDepressedNondepressedUnadjusted odds ratio (95% CI)
  1. *Students n test.

  <200.7%0.6%1.4 (0.3–6.0)
  20–2415.7%11.5%1.6 (1.1–2.4)
  25–2932.5%38.8%1.0 (reference)
  30–3434.6%34.8%1.2 (0.9–1.6)
  >3416.4%14.2%1.4 (1.0–2.1)
  Mean age29.729.6P= 0.69*
  046.1%53.1%1 (reference)
  1–247.5%44.9%1.2 (0.9–1.6)
  >26.4%2.0%3.6 (2.1–6.2)
Marital status   
  Solitary5.2%2.5%2.2 (1.2–3.9)
  Married/cohabiting94.8%97.5%1 (reference)
Working status   
  Welfare payments17.6%6.0%3.6 (2.5–5.0)
  Unemployment relief17.9%15.3%1.4 (1.0–2.0)
  Working/ sick leave/ student64.4%78.7%1 (reference)
Table 2.  Psychopathological status and social support by postpartum outcome. Values are given as %, unless otherwise stated. GHQ = General Health Questionnaire.
VariableDepressedNondepressedUnadjusted odds ratio (95% CI)
  1. *Levels of psychological distress: GHQ score 4 = high; GHQ score 2–3 = moderate; GHQ score < 2 = low.

  2. Mann- Whitney U test.

Positive history of psychiatric disease19.6%7.3%3.1 (2.3ndash;4.2)
Family history of psychiatric disease20.2%14.0%1.6 (1.1ndash;2.1)
Level of psychological distress, third trimester*   
  High45.2%13.9%7.3 (5.4ndash;9.9)
  Moderate25.1%19.3%2.9 (2.1ndash;4.0)
  Low29.7%66.8%1 (reference)
  Mean score of GHQ3.981.46P= 0.003
Alcohol intake, third trimester (glasses per week)   
  ≥534%2.3%1.3 (0.7ndash;2.6)
  1ndash;522.8%31.3%0.7 (0.5ndash;0.9)
  <173.8%66.4%1 (reference)
Present or prior drug abuse2.1%1.3%1.7 (0.7ndash;3.9)
Perceived social isolation   
  Always10.6%2.0%9.0 (5.5ndash;14.6)
  Occasionally68.1%62.7%1.8 (1.3ndash;2.4)
Never21.2%35.3%1 (reference)
Table 3.  Perinatal events by postpartum outcome.
VariableDepressedNondepressedUnadjusted odds ratio (95% CI)
Not attending parenting classes52.8%40.2%1.7 (1.3–2.2)
Prior induced abortion18.6%18.7%1.0 (0.7–1.4)
Mode of delivery   
  Acute caesarean section6.2%5.5%1.1 (0.7–1.9)
  Instrumental delivery12.8%13.0%1.0 (0.7–1.4)
  Spontaneous delivery81.0%81.5%1 (reference)
Apgar.score< 7 by 5 min0.4%0.8%0.4 (0.1–3.1)
Preterm delivery (gestational week < 37)6.1%3.8%1.6 (1.0–2.7)
Low birthweight (< 2500 g)4.3%3.1%1.4 (0.8–2.5)
Multiple pregnancy1.5%2.0%0.7 (0.3–2.0)
Pregnancy complications   
  Vaginal bleeding3.6%2.5%1.5 (0.8–2.9)
  Premature contractions4.4%3.7%1.2 (0.6–2.1)
  General physical discomfort0.7%0.4%1.8 (0.4–8.1)
  Pelvic.joint complaints6.9%5.8%1.2 (0.7–2.0)
  Sick leave due to pregnancy   
  Any complications6.6%4.6%1.5 (0.9–2.4)
  No complications77.7%83.1%1 (reference)
Admission to hospital due to pregnancy complications13.5%7.4%1.9 (1.4–2.8)
Puerperal complications7.7%7.4%0.8 (0.5–1.3)
Neonatal complications   
  Child admitted to neonatal department10.6%8.9%1.2 (0.8–1.8)
Congenital malformations2.5%1.0%2.6 (1.2–5.7)
Place of delivery   
  Hospital99.6%98.6%0.3 (0.1–1.9)
  Home0.4%1.4%1 (reference)
Postnatal hospital stay (days)   
  <519.4%15.4%1.3 (1.0–1.8)
  2–529.1%30.3%1 (0.8–1.3)
  <151.5%54.3%1 (reference)
Never established5.4%2.4%2.4 (1.7–4.1)

Subsequent multivariate associations are shown in Table 4, including the variables for both antenatal and perinatal risk factors. These results show that psychological distress is the variable with the strongest association with postpartum depression with an odds ratio of 6.3 (4.4–9.1) for the most distressed group of women, compared with women in the low distress category. Social isolation showed on multivariate level a markedly lower but still strong association with postpartum depression with an odds ratio of 3.6 (1.9–7.0), compared with women who did not report feelings of social isolation. Among other variables, only high parity and prepregnant history of psychiatric disease were significantly associated with postpartum depression increasing the women's risk three- and twofold respectively. It is notable that no perinatal event or paediatric complications remained significant risk factors when applied into multivariate analysis.

Table 4.  Significant adjusted odds ratio (95% CI) n= 3546. OR = odds ratio; GHQ = general health questionnaire.
 Antenatally adjusted OR (CI)*Perinatally adjusted OR (CI)*
Antenatal information  
  <201.2 (0.2–5.8)1.1 (0.2–5.6)
  20–241.3 (0.8–2.1)1.3 (0.8–2.1)
  30–341.0 (0.7–1.5)1.0 (0.7–1.5)
  >341.1 (0.7–1.8)1.1 (0.7–1.8)
>23.8 (1.8–8.0)2.9 (1.3–6.3)
Marital status  
  Single1.1 (0.5–2.4)1.1 (0.5–2.4)
Occupational status  
  Welfare payments1.7 (1.0–2.8)1.5 (0.9–2.6)
  Unemployment relief1.2 (0.8–1.8)1.1 (0.8–1.7)
  Working/ student/ sick leaveReferenceReference
Past history of psychiatric disease  
  Yes2.1 (1.4–3.2)2.1 (1.4–3.2)
Family history of psychiatric disease  
  Yes1.4 (1.0–2.1)1.5 (1.0–2.1)
Perceived social isolation  
  Always3.6 (1.9–7.0)3.6 (1.9–7.0)
  Occasionally1.5 (1.1–2.2)1.5 (1.1–2.2)
Level of psychological distress during pregnancy  
  High6.3 (4.4–9.1)6.1 (4.2–8.9)
  Moderate3.2 (2. 1–4.7)3.1 (2.1–4.7)
Perinatal events  
Not attending parenting classes 1.5 (1.0–2.1)
Admitted to hospital during pregnancy 1.2 (0.7–2.0)
Congenital malformations 1.2 (0.3–4.5)
Breastfeeding never established 1.5 (0.7–3.1)
*Column variables adjusted for all variables in column.
Levels of psychological distress: GHQ score 4 = high; GHQ score 2–3 = moderate; GHQ score < 2 = low.

The two indices performed almost identically with an area under the receiver-operator curve of 0.72 for antenatal index, and 0.71 for the combined antenatal and postnatal index (Fig. 1) and a maximum positive predictive value of about 30%, assuming a prevalence of 5.5% (Fig. 2).

Figure 1.

Receiver-operator characteristic curve for antenatal and perinatal indices.

Figure 2.

Positive predictive value for antenatal and perinatal indices at different cut off scores.

Analyses of nonrespondents showed that women who did not complete the Edinburgh Postnatal Depression Scale had a significant higher frequency of especially psychosocial risk factors than women who had completed the Edinburgh Postnatal Depression Scale (Table 5).

Table 5.  Differences in responders and nonrespondents to the Edinburgh Postnatal Depression Scale.
VariableNonrespondents (%)Responders (%)P
Young age (20–24)
High parity (> 2)
Low working status22.26.70.0001
Past history of psychiatric disease8.48.00.58
Family history of psychiatric disease13.314.30.38
Lack of confidence4.32.40.002
Psychological distress (high level)
Attending parenting classes39.959.00.0001
Admission to hospital due to pregnancy complications9.77.70.01
Congenital malformations1.61.10.08


To our knowledge this is the largest study of postpartum depression. The systematic, prospectively collected information about potential risk factors among a nonselected population and use of a standardised and well validated screening instrument provided the possibility of estimating the prevalence of postpartum depression in a general obstetric population. In addition, the predictive power of some uncommon potential risk factors was estimated.

The point prevalence of 5.5% with postpartum depression seen in this study is slightly lower than that observed in other Scandinavian studies with comparable study populations45,46. Thus, a prevalence of approximately 7% using a similar cut off score of 12/13 on the Edinburgh Postnatal Depression Scale has been reported.

This may in part be explained by differences in the timing of assessment. In our study prevalence was assessed at four months postpartum, and previous studies have found an increased rate of onset in the first three months following delivery10. Thus, an earlier assessment could have found a higher prevalence8.

The low point prevalence may also be influenced by selection bias due to the high proportion of nonrespondents. This is supported by the fact that nonrespondents had a higher frequency of risk factors, compared with women who had completed the Edinburgh Postnatal Depression Scale.

With these limitations in mind, our study and previous Scandinavian studies suggest that the prevalence of postpartum depression in a nonselected Scandinavian population is lower than reported from non-Scandinavian countries34,47–49. Our study further supports previous evidence that psychosocially disadvantaged women are at high risk of developing postpartum depression. The strongest antenatal predictor was psychological distress. This indicates that postpartum depression, at least in some women, could be viewed as a continuum of depressive symptoms, deteriorating over time. The other strong risk factor associated with postpartum depression was social isolation.

The strong association between lack of social support and postpartum depression has been demonstrated previously in a prospective study in a community-based study49: women who did not attend parenting classes were at higher risk of subsequent postpartum depression than those who attended classes. One earlier study has confirmed this finding50.

Our results do not support the suggestion that pregnancy or delivery complications per se should represent any cause for the development of postpartum depression. Thus, no obstetric risk factor could be identified on the multivariate level among the extensive amount of perinatal information collected.

These findings contradict previous studies concerning the role of perinatal events, especially caesarean section23,24,51–53. Our study includes both emergency and elective caesarean section, since emergency caesarean section may constitute a more traumatic experience. To facilitate comparisons with other studies, elective and emergency caesarean section were also combined, but remained unassociated with postpartum depression.

The design of the present study allowed for testing the predictive power of the identified risk factors on a sample not used for extracting the predictive variables.

As seen in Figs. 1 and 2, the maximum positive predictive value of 0.30 corresponds to a sensitivity and specificity of 0.79 and 0.50, respectively. Psychological distress and social isolation were by far the strongest risk factors for postpartum depression. Thus, the results suggest that with an estimated prevalence of 5.5%, approximately one out of three women with severe psychological problems and social isolation in late pregnancy after delivery will develop postnatal depression. The antenatal prediction is not likely to be improved significantly by incorporating perinatal parameters. As seen, these factors did not improve the predictive performance (Figs. 1 and 2).

The present study does not take into account possible aetiological factors operating from delivery and up to the time of assessment four months postpartum. Part of the demonstrated lack of precision of our prediction could thus be attributed by such potential factors. It cannot be ruled out that factors other than the ones investigated may be able to improve the detectability; however, we have included and tested all relevant risk factors known from the literatute.

To our knowledge only one prior study has tested antenatal prediction of postpartum depression on a larger sample34. However, in contrast to the previous study, we have assessed the predictive power of psychological morbidity and lack of social support in relation to postpartum depression using a standardised and well validated instrument.

Many screening questionnaires have been developed to assess psychosocial factors. The General Health Questionnaire is, in our opinion, the best validated and most widely used instrument to assess psychological morbidity and could in its short 12-item version prove to be an important antenatal care tool for detecting women at risk for postpartum depression. The fact that the condition is seldom identified in the postpartum period54 stresses the importance of antenatal strategies targeted at the high risk group.

Twenty percent of severe postnatal depression become chronic and delay of treatment is the greatest factor in the development of chronic depression55. Thus, special attention should be given to women at risk to ensure early identification and adequate treatment. Our findings suggest that screening for psychological morbidity in late pregnancy, as well as focusing on reported perception of social isolation may be useful in identifying women at risk antenatally who may develop postnatal depression.


This study was supported the Ministry of Health (The National Health Fund for Research and Development) and Eli Lilly (Denmark). The authors would like to thank the staff at the Perinatal Epidemiological Research Unit at the University Hospital in Aarhus for their assistance during data collection.