The Role of Social Networks and Support in Postpartum Women's Depression: A Multiethnic Urban Sample
- First Online:
- Cite this article as:
- Surkan, P.J., Peterson, K.E., Hughes, M.D. et al. Matern Child Health J (2006) 10: 375. doi:10.1007/s10995-005-0056-9
- 561 Views
Objectives: This study examined the relationship of social support, and of social networks, to symptoms of depression in a multiethnic sample of women having recently given birth. Methods: Women at community health centers in a Northeastern city were randomly sampled from groups stratified by race/ethnicity (African American, Hispanic, and White) and postpartum interval. Mother's score on the Center for Epidemiologic Studies of Depression Scale (CES-D) was the dependent variable. Main independent variables included the Medical Outcomes Study (MOS) Social Support Survey and a social network item. Univariate statistics assessed the relationship between CES-D score and each of the independent variables. Multivariate linear regression models included core sociodemographic variables alone, the core model with each of the social support and social network variables added separately, and all variables together. We evaluated interactions between race and social support, race and social networks, and social support and social networks. Results: The multivariate models with MOS Social Support and core variables indicated that each 10-point increase in the MOS Social Support Survey was related to a 2.1-unit lower score on the CES-D (95% CI −2.4, −1.7). The inclusion of the social network variable into the core model showed that having two or more friends or family members available was associated with a 13.6-point lower mean score on the CES-D (95% CI −17.5, −9.6), compared to women reporting none or only one available person. Conclusions: Both social support and social networks were statistically significant and independently related to depressive symptomatology.
KEY WORDS:postpartumdepressionsocial supportsocial networksmultiethnic sample.
Depression following the birth of a child is common, with prevalence estimates ranging from 10 to 20% (1). In response to a single question, up to 65% of women describe being slightly, moderately, or very depressed at 6 months after delivery (2). Depression in the postpartum period has been considered distinct from other types of depression since it can be brought on by hormonal and biological changes related to childbirth (2). Defined in the Diagnostic and Statistical Manual IV by onset during the first 4 weeks after delivery (3), postpartum depression is typically assessed within the first year after a birth (4–6). It is considered an affective disorder of moderate to severe symptoms (7) on the continuum between postpartum blues and psychosis (5, 8).
Postpartum depression is associated with marked distress and dysfunction (9) and increased risk of subsequent depression (10). Mood disturbance and impairments are characteristic of depression at any stage of life (3). In addition, the consequences of postnatal depression extend beyond the mental health of the mother to her relationship with the child (11, 12). Parenting behavior can be compromised (11) and infants of depressed mothers show lower likelihood of secure attachment (12). Through its effects on parenting, depression has been linked to children's adverse cognitive and motor development (13).
Previous research has shown that depression is associated with the demands of childrearing after delivery (14) and an inverse association with social support (15–17), a potentially modifiable factor. Social support and depression are more strongly related at 2 months after delivery compared to at 2 weeks, which may be attributable to the fact that later depression is more psychologically and less physiologically determined (16). Depressive symptoms at 9 and 12 months postpartum have been related to a poor marital relationship (18), which may suggest that a good quality spousal relationship may buffer stress and provide social support.
Social support may be particularly important among low-income parents. Women living in poverty have approximately double the prevalence of stressful life events and depression compared to nonpoor women (19). Hashima and Amato in 1994 (20) found a negative association between parents’ punitive behavior of children under the age of 5 and social support among families living in poverty, but no association among higher income families. Furthermore, because immigrants can face increased stressors related to discrimination or the stress of adjusting to a new culture, social support might be particularly relevant in that context. Because discrimination fosters and maintains inequalities, Belle and Doucet (21) suggest that a stress response to the experience of racism may influence subsequent coping strategies. In addition to studying the relationship of social support to postpartum depression, our study sought to examine the relationship of social networks to symptoms of depression in a multiethnic sample of women who have recently given birth.
The Postpartum Women's Health Study was conducted at community health centers in the greater metropolitan areas of a Northeast City. These health centers serve a low and moderate income, racially and ethnically diverse urban population. This cross-sectional study was reviewed and approved by the human research committees of Brigham and Women's Hospital, Children's Hospital, and the Harvard School of Public Health, as well as by the policy boards of each of the participating health centers.
Women who delivered a live, singleton infant between May 1996 and May 1998 were selected from health center obstetrical and pediatric patient lists. The sample was stratified by race/ethnicity (African American, Hispanic, and White) and postpartum interval (6 weeks to 6 months, over 6 months to 12 months, over 12 to 18 months, and over 18 months to 24 months). Women were randomly sampled to reach a quota of 60 in each of the 12 strata. For three strata in which there were fewer than 60, all were selected. Maternal-related exclusion criteria included current pregnancy, major physical disability such as blindness or deafness, severe chronic mental illness such as schizophrenia or bipolar disorder, or current chemotherapy treatment. Participants were required to be able to read in either English or Spanish. Infant-related exclusion criteria included infant death, birth outside of the United States, severe medical condition requiring frequent hospitalization or in-home medical personnel, and the infant not living with his/her biological mother. Women who had given birth to a second infant at time of selection were reassigned to a stratum based on the age of the younger child.
A research assistant sent recruitment letters to potential participants who were given a 2-week period to return a card declining participation. The research assistant then contacted the women who did not return the card. Extensive attempts were made to locate all those who passively consented to be contacted by telephone.
Using health center patient lists, 1330 women were identified. Of these, 145 were ineligible according to exclusion criteria. An additional 39 were known to have left the state or the country. Despite extensive efforts, 271 women could not be located, 298 no longer had a child under the age of 2 by the time we were able to locate them and schedule a survey date. A total of 577 women were successfully located with adequate time to schedule a survey date. Of these, 415 agreed to participate, yielding a 49% response rate (of eligible and potentially contactable women) and a 72% response rate of those we succeeded in contacting.
We created a composite questionnaire comprising general health status and mental health measures (SF-36), postpartum specific measures, and measures of social support and social networks. The dependent variable was the mother's score on the Center for Epidemiologic Studies of Depression Scale (CES-D) (22) of depressive symptomatology during the last week. This was used as a continuous variable. Main independent variables included the score on the Medical Outcomes Study (MOS) Social Support Survey (23) and number of people mentioned in the social network items (24). Questions from the MOS Social Support Survey tap into a variety of subdomains of social support and include items such as having someone to give advice, confide in, listen to you, etc. The MOS Social Support Survey is a 20-item instrument with a continuous score ranging from 0 to 100 points, with a higher score representing more support. The instrument is widely used and considered to have high reliability and validity (23, 25). Information about social networks included questions asking how many friends and how many relatives one feels close to (24). These questions were added to create one variable representing the number of network connections including friends or relatives. Because we hypothesized that having many people to get support from would be very different from the experience of having nobody or only one person available and because there were only seven respondents reporting no support person, we categorized social networks into categories of 0–1 or 2 or more.
Comprehensive sociodemographic and questions assessing acculturation were taken from the Breast Cancer Core Questionnaire (26). English and Spanish language versions were developed. Standard translations were used where available. Translations of additional items developed for this survey were prepared by a bilingual/bicultural medical interpreter fluent in Caribbean Spanish dialects spoken by the study population, and back-translated and reviewed by two additional native speakers.
Following informed consent, questionnaires were completed at a health center, the participant's home, or other location chosen by the participant. A trained research assistant was present to provide assistance, and to screen and refer to services in cases where responses to questions indicated risk for violence or self-inflicted harm. Participants were paid $15.
Using SAS (Statistical Analyzing Systems, Version 8.1, Cary, NC), univariate statistics were performed to assess the relationship between CES-D score and each of the independent variables. Demographic variables which were included in the model regardless of their univariate significance included race/ethnicity (African American, Hispanic, White), mother's age (continuous), and educational attainment (less than high school graduate/high school graduate or more). Variables which attained a univariate significance of p ≤ 0.15 for Pearson correlations, t tests, trend tests, and ANOVA where appropriate were subsequently entered into a series of multivariate linear regression models with other variables with which they were considered to be closely related. These categories were determined a priori and include 1) family's financial status, 2) insurance status, 3) acculturation, and 4) social networks and social support. Associations with more stringent p values of ≤0.10 in intermediate models were retained for further analysis. Sociodemographic variables were tested only for their univariate significance before determining whether they would be entered into the model. These variables included number of children under the age of 5 (1, 2, 3); time since pregnancy (6 weeks to 6 months, over 6 months to 12 months, over 12 to 18 months and over 18 months to 24 months), stable housing (yes/no); car availability (yes/no); experience of discrimination based on gender, race, or social class (yes/no), and crowding. The variable representing crowding was created by dividing the number of people living in the house by the number of rooms, after which categories were made representing the average number of people per room (less than one, one, or more than one).
After a core multivariable model was constructed through this process, additional linear regression models were created by adding each of the MOS Social Support and the social network variables separately and simultaneously to this core model to examine their independent contribution to variance in the CES-D score. Finally, we expanded the model with all variables to test interactions between race/ethnicity and social support, race and social networks, and social support and social networks.
Because we found no significant evidence that the associations of interest varied by race/ethnicity or postpartum interval, we do not report stratified results.
Mean CES-D Scores and Demographic Characteristics Among Multiethnic Postpartum Women
CES-D score (mean, SD)
Univariate p value
High school graduate or more
Less than high school graduation
Mothers age, (year), mean (SD)
Time postpartum (month), mean (SD)
0.52 (trend test)
Children <5 years of age
0.05 (trend test)
<0.01 (trend test)
Ever felt discriminationc
Yes—at least one kind of discrimination
No—reports no discrimination
0.03 (trend test)
$50,000 or more
Income from job
Income from spouse's job
Income from family member's job
CES-D score (mean, SD)
Univariate p value
Income from social security
Income from unemployment
Aid to families of dependent children
Hard time making ends meet in last 3 years
Self or free
Medicaid or healthy start
Insurance from work
Mother speaks English
Mother's age at immigration, mean (SD)
Length of time in U.S. (years), mean (SD)
One or both parents not U.S. born
Support and networks
Living with a partner
MOS Social support survey, mean (SD)
Social network (family or friends available)
One-quarter reported experiencing some kind of discrimination based on race, class, or sex. Half the sample had at least one parent born abroad (52%), and 41% had immigrated themselves. Although 36% of participants did not report living with a partner, almost 95% reported having support of two or more social contacts. The average score (SD) on the MOS Social Support Survey was 76.9 (24.9).
Mean CES-D scores and tests of univariate association are displayed in the final two columns in TableI. African Americans and Hispanics had higher CES-D scores than Whites on the CES-D (mean (SD): African American 14.2 (10.8); Hispanic 15.2 (10.3); White 11.4 (9.2)). Education and income were inversely related to scores indicating more depressive symptoms. Those participating in means-tested programs who were uninsured and who reported having a hard time making ends meet had higher CES-D scores. Less acculturation seemed to be associated with higher depressive symptomatology; those with non-English-speaking mothers and who had at least one foreign-born parent had significantly higher CES-D scores. Finally, women with lower scores on MOS Social Support Survey, who did not live with a partner and had nobody or only one family member or friend available, had significantly higher CES-D scores in univariate analyses.
Linear Regression Models of Variables Associated with Depressive Symptomatology (CES-D) Among Multiethnic Postpartum Women
Model 1: core model (n = 398)
Model 2: core model with social support (MOS) (n = 397)
Model 3: core model with social networks (n = 392)
Model 4: core model with social support (MOS) and networks (n = 391)
Age of mother (per 10-year increase)
Education high school/< high school (ref)
Income from spouse's job yes/no (ref)
Problems with money in last 3 years yes/no (ref)
Number of children under 5 years of age (1, 2, 3)
Discrimination any/no discrimination (ref)
Social support questions from MOS (per 10-point increase)
Social networks 2 or more friends or family members available/0–1 available (ref)
The multivariate models with MOS Social Support and core variables indicated that each 10-point increase on the MOS social support was related to a 2.1-unit lower score on the CES-D (95% CI −2.4, −1.7) (Model 2). The inclusion of the Social Network variable into the core model showed that having two or more friends or family members available was associated with a 13.6-point lower mean score on the CES-D (95% CI −17.5, −9.6), compared to women reporting none or only one available person (Model 3). In spite of the correlation that might be expected between social support and social networks, both of these variables were statistically significant and independently related to depressive symptomatology. When all variables were adjusted for simultaneously (Model 4), having two or more friends or family members available was associated with a 9.4-point lower mean score on the CES-D, compared to women who report having zero or one available friend or family members (95% CI −13.1, −5.8). In this model each 10-unit increase on the MOS Social support scale was associated with a 1.8 lower mean CES-D score (95% CI −1.5, −2.2) (Model 4). When the social support and social network variables were simultaneously entered into the model all the other associations were attenuated and became nonsignificant except for income from spouse's job. Although income from spouse's job remained significant, the effect estimate changed from −4.8 in the model without social support or social networks to −2.8 in the model including both of these variables. None of the interactions between social support and networks, or race/ethnicity with either social support or social networks were significant.
Our study indicated that both social networks and social support were independently and inversely related to depressive symptomatology. A similar study with a longitudinal design would provide insight into changes over time and could provide more information regarding causality. Women who reported having two or more available friends or family members showed notably fewer depressive symptoms, which corresponded to approximately a 9-point lower CES-D score, compared to women reporting zero or only one available friend or relative. The association between social support and depressive symptomatology corresponded to around a 2-point change in mean CES-D score, for each 10-unit increase in the mean score on the MOS Social Support Survey. Our findings underscore an independent contribution of social network availability and social support to lower depressive symptomatology.
In terms of social support, the inverse relationship between depressive symptoms and social support is consistent with prior research (16–18). We believe that our social networks measure taps into the availability of sources of support, while the MOS Social Support Survey measures the quality of the support from these social network contacts. Prior research has found the number of people providing emotional support 3 weeks postpartum and the number of people providing informational support at 6 months postpartum to be related to depressive symptoms at 6 months postpartum (27). In contrast to our findings, a study of adolescent mothers found no differences between nondepressed and mild to severely depressed groups in terms of how frequently they reported receiving social support. Their social networks measure, ‘‘conflicted networks,’’ which assessed the number of people who were a source of support and conflict during the past month, also did not differ between the groups (28). Brugha et al. (29) measured negative interaction with relatives, friends, and other contacts (such as work colleagues and acquaintances), but did not find it to be a significant predictor of depressive symptoms. In our study, tests for interactions between racial group and social support and between racial group and social networks provided no evidence that the influence of social support or of social networks on depression varied by race.
In the core model unadjusted for social support or social networks, variables we found related to depressive symptoms included discrimination, education, income from spouse's job, problems making ends meet in the last 3 years, and number of children under the age of 5. Prior research has identified economic hardship as a correlate of depressive symptoms (27, 30). Low-income women face a high number of economic and other stressors (31) that may contribute to depression. To explain associations between number of children and postpartum depression (29, 32), some have suggested that women with more children may have heavier demands and responsibilities (32). Our data support the previously documented inverse relationship between depressive symptoms and educational level (29, 32). Although we identified income from spouse's job as associated with postpartum depressive symptoms in all of our models, having problems with money in the last 3 years was related to more depressive symptoms in both the core model and in the model including social networks. Finally, our study indicated that women who reported any kind of discrimination were more likely to have depressive symptomatology. Although discrimination has been linked to depression in a variety of immigrant (33, 34) and workplace settings (35), as far as we know it has not been previously studied among women during the first 2 years after delivery.
This study provided the opportunity to examine correlates of depression of recent mothers in a racially, ethnically, and economically diverse population, including an approximately equal number of African Americans, Hispanics, and Whites. Collection of data in both Spanish and English allowed for the participation of recent Hispanic immigrants to the United States. Because of its simplicity, the use of a single social networks question asking how many friends and relatives are available lends itself well as a convenient and practical tool to the clinical setting, where time is often limited.
The cross-sectional design of our study limits inferences about the direction of the association between social support and depressive symptoms. The fact that the sample was selected from health centers located in particular neighborhoods, rather than being population-based, restricts the generalizability of the study to women who are self-selected to use these community health centers. However, further analysis of our data showed that nonrespondents did not differ from respondents on postpartum interval, race/ethnicity, or health center site (data not shown). Given the high validity of the CES-D and MOS social support instruments the potential for nondifferential misclassification would be low.
Our findings indicate that social networks and social support are important factors related to postpartum women's mental health and draw attention to the potential utility of screening for them in community health centers for postpartum women. Questions on social networks and social support, in combination with review of family economic resources, could be integrated by primary-care practitioners to screen for risk factors for depression among mothers of infants and children under 2 years of age. The concise nature of our social network questions, in the context of time constraints health centers often face, is convenient for this setting. Further research may be needed to develop a social support screening measure, similar to the MOS Social Support Survey, which is more succinct.
This study was supported by the Maternal and Child Health Bureau (MCJ-259356). The authors express their appreciation to Dr. Marie McCormick, whose expertise guided this project from its inception. We also gratefully acknowledge the contributions of Rachel Levine and Cary Perry, who served as project coordinators and research associates, and Peter Gaccione for his contribution to data analysis.