Maternal Patterns of Antenatal and Postnatal Depressed Mood and the Impact on Child Health at Three Years Postpartum

Abstract

Objective

The consequences of maternal depressed mood on children’s growth, health, cognitive and language development are examined over the first three years of life.

Method

Pregnant women in 24 peri-urban township neighborhoods in Cape Town, South Africa (N=1,238 mothers) were randomized by neighborhood to a home visiting intervention or a standard care condition. Reassessments were conducted for 93%-85% of mothers at two weeks, 6, 18, and 36 months post-birth. Regressions were conducted on measures of children’s growth, behaviour, language, and cognition to examine the impact of four patterns of depressed mood: antenatal only (n=154, 13.8%), postnatal only (n=272, 24.3%), antenatal and postnatal (n=220, 19.7%), and no depressed mood on any assessment (n=473, 42.3%).

Results

Patterns of depressed mood were similar across intervention conditions. Depressed mothers were significantly less educated, had lower incomes, were less likely to be employed or to have electricity; were more likely to report problematic drinking of alcohol, experience food insecurity, interpersonal partner violence, and to be HIV seropositive. At 36 months, the pattern of maternal depressed mood over time was significantly associated with children’s compromised physical growth, both in weight and height, and more internalizing and externalizing symptoms of behaviour problems. Measures of language and cognition were similar across maternal patterns of depressed mood.

Conclusions

Mothers who report depressed mood face significantly more life challenges, both environmental stressors related to poverty and other problematic behaviors. More proximal, postnatal depressed mood appears to have a larger influence on their children, compared to antenatal depressed mood.

Introduction

Depressed mood is a significant global challenge facing millions of women and mothers annually. Meta-analytic reviews document the prevalence of antenatal depression globally at 15.6% and an additional 19.8% of mothers are depressed postnatally (Coll et al., 2017; S. H. Goodman et al., 2011; World Health Organization, 2004). The negative consequences of maternal depressed mood have been well-established in high income countries (HICs) (R.E. Black et al., 2008; Victora et al., 2008) and is emerging in low and middle income countries (LMICs) (R. E. Black et al., 2013; Wachs, Black, & Engle, 2009). Three studies in Asian countries have demonstrated a strong relationship between maternal antenatal depression and children’s retarded growth (Nasreen, Kabir, Forsell, & Edhborg, 2011; Patel, DeSouza, & Rodrigues, 2003a; Rahman & Creed, 2007; Rahman, Iqbal, Bunn, Lovel, & Harrington, 2004; Rahman, Malik, Sikander, Roberts, & Creed, 2008). The evidence is more mixed in Africa with a number of studies finding no relationship between child growth and antenatal depression (Tomlinson, Cooper, Stein, Swartz, & Molteno, 2006; Verkuijl et al., 2014).

The literature in LMIC typically assesses depression at one time point, usually post-birth. It is critical that we observe maternal depression longitudinally over time, concurrent with children’s adjustment. To date, the majority of longitudinal studies have only been conducted in HICs, where there is substantial evidence of how the longer mothers are depressed, the worse their children’s adjustment (Parsons, Young, Rochat, Kringelbach, & Stein, 2012). However, the impact of multiple depressive episodes during the developmental periods of pregnancy and early childhood is poorly understood (G. Dawson et al., 2003).

Disentangling the differential impacts of ante- and postnatal depression is difficult, but growing evidence suggests that the differences are important (including on child adiposity, see Ertel, Koenen, Rich-Edwards, & Gillman, 2010). A study by Deave, Heron, Evans and Emond (2008), for instance, found an association between persistent maternal depression during pregnancy and later developmental delays in their children. While the authors note that postnatal depression has a modifying effect of children’s developmental delays, they find an independent and consistent association between antenatal depression and child development (Deave, et al, 2008). There is reason to believe that there may be differential impacts, depending on when a mother is depressed. When mothers are depressed during the antenatal period, fetal development may be affected (as well as subsequent development); and when mothers are depressed postnatally, depressed mood can influence neonatal and early child development.

Due to its timing, antenatal depression is likely to affect the child’s birth status and biology. Antenatal depression and low birth weight and fetal growth restriction are significantly correlated (Evans, Heron, Patel, & Wiles, 2007; Wisner et al., 2009). Pregnant women with antenatal depression are also more likely to suffer from obstetrical complications such as preeclampsia (Abedian, Soltani, Mokhber, & Esmaily, 2015; Field et al., 2004). The risk of preterm delivery increases significantly when mothers are antenatally depressed, and preterm delivery contributes to later cognitive, behavioral and emotional problems in children (Luoma et al., 2001). Antenatal depression also exposes the neonate to elevated maternal cortisol (Field et al., 2004). Elevated levels of cortisol, in turn, negatively influence infant temperament, and have been linked to childhood psychopathology (Davis et al., 2007).

Regardless of whether it is followed by an episode of postnatal depression or not, antenatal depression negatively effects child outcomes (Deave et al., 2008; Halligan, Murray, Martins, & Cooper, 2007). Each of these depression-linked birth and physiological sequelae may then have implications for the child’s subsequent development. One of the factors determining these sequelae depends on whether the mothers’ depression remits or not. If it does, the child may have a biological and physiological architecture that put them at risk for subsequent problems. Although antenatal depression predicts postnatal depression, antenatal depression may be more common than postnatal depression (Evans, Heron, Francomb, Oke, & Golding, 2001; Llewellyn, Stowe, & Nemeroff, 1997), and the children of antenatally depressed mothers require separate analysis.

Postnatal depression is also significantly associated with negative child outcomes (Murray, FioriCowley, Hooper, & Cooper, 1996; Richter, Griesel, & Barbarin, 2000). Maternal depression can affect maternal child rearing behaviors and maternal sensitivity, via decreased maternal availability and suboptimal parenting. Ultimately, examining children separately, depending on whether the child is exposed to ante- or post-natal maternal depression, may contribute to our capacity to disentangle ‘programming’ effects from subsequent ‘socioemotional’ effects. This is important given that recent work has shown how some effects on child development attributed to postpartum depression are caused, in part, by antenatal depression (Deave et al., 2008). Finally, it is worth attending to children exposed to the synergistic effects of antenatal and postnatal (and continuing) maternal depression (programming plus socioemotional effects, as it were). It is plausible that the children of chronically depressed mothers may have the worst outcomes.

Therefore, with longitudinal observations, it is possible to classify mothers as antenatally depressed only, postnatally depressed only, depressed both antenatally and postnatally or not depressed. These categories have been observed by other researchers. Two recent articles with large data sets from the United States, a high-income country, classified mothers in a similar way to the one we used in this article (Mora et al., 2009; Sutter-Dallay, Cosnefroy, Glatigny-Dallay, Verdoux, & Rascle, 2012). Our current study allows us to examine this classification system among women in South Africa.

Mothers with depressed mood may differ from their peers

We expect that maternal depression is related to a number of structural stressors in mothers’ lives. Poverty has consistently been associated with higher rates of mental health symptoms, especially depression (Lund et al., 2010). Low social status, a lack of social support, and conflictual partnerships are also associated with perinatal maternal depression (R. E. Black et al., 2013; O’Hara, 2009). Education and marriage both confer social status and have been linked to fewer mental health symptoms and disorders (Simon, 2002). Many South African women live with their family of origin, even when they have sexual partners – the financial costs of marriage and an inability to pay the penalties levied for young men getting a women pregnant outside of marriage result in many women living with their family of origin (Montgomery, Hosegood, Busza, & Timaeus, 2006). Thus, partnerships are often unstable and financial instability is often a way of life for families (O’Hara, 1996).

In South Africa, however, there are additional risks facing all mothers. South Africa has the highest number of HIV persons of any country globally: 26% of pregnant mothers are living with HIV. More than 98% of women get tested for HIV in pregnancy and, therefore, many pregnant women are likely to learn of their HIV status in pregnancy (Joint United Nations Programme on HIV/AIDS, 2014; le Roux et al., 2011). Mothers living with HIV have been found to have higher rates of depression (Dyer, Stein, Rice, & Rotheram-Borus, 2012; le Roux et al., 2013; Rotheram-Borus, Tomlinson, Le Roux, & Stein, 2015). However, the reports of depression are typically assessed around the time of HIV diagnosis in pregnancy. The duration of maternal depression among mothers living with HIV is less clear. The depressed mood of mothers living with HIV may be limited to the time of HIV diagnosis.

In addition to HIV, South Africa has the highest per capita rate of alcohol abuse (World Health Organization, 2014) and the highest documented rate of Fetal Alcohol Spectrum Disorder (May et al., 2014). Women live in communities permeated by shebeens (alcohol bars) and have partners who drink daily (Eaton et al., 2012), with one in four women also drinking alcohol prior to knowing that they are pregnant (O’Connor et al., 2011). Alcohol is itself a depressant and may be associated with more frequent reports of depressed mood (Cronce & Larimer, 2013). Thus, there may be differences in which mothers become depressed in LMIC which are different from issues experienced by depressed mothers in high income countries.

The impact of depressed mood on children

Physical growth is the primary area linked to maternal depression in LMIC (Parsons, Young, Rochat, Kringelbach, & Stein, 2011; Patel, Rahman, Jacob, & Hughes, 2004; Rahman et al., 2009). More than 40% of township households face food insecurity (Tsai, 2013). Food insecurity is most typically reflected in children’s low birth weight (i.e., < 2500 grams), stunting and malnutrition (Wachs et al., 2009). As noted above, however, the influence of maternal depressed mood on children’s growth have been mixed in the South African studies (Tomlinson et al., 2007; Verkuijl et al., 2014).

There is substantial evidence that maternal depressed mood negatively impacts children’s development in high income countries (Bauer et al., 2015; O’Connor et al., 2014; Stein et al., 2014), with between 10-15% of children’s emotional and behavioral adjustment attributable to maternal depression (Glover, 2014). Similarly, cognitive deficits have also been documented in a series of reviews in high-income countries, associated with maternal depression (Apter-Levy, Feldman, Vakart, Ebstein, & Feldman, 2013). The cognitive deficits are primarily measured as IQ scores, but have typically not been studied in LMICs. Neither of these domains, behavioral adjustment and cognition, have been studied in LMIC. This study is one of the first to examine the impact of maternal depressed mood on these domains of child functioning.

Interventions may mediate the impact of maternal depressed mood

This study examines patterns of maternal depression over the first three years of life, as well as the influence of mothers’ depressed mood on their children. These analyses are conducted within the context of a randomized controlled trial on the efficacy of home visiting. It is critical to examine both the effects of home visiting, as well as the concurrent influence of maternal depressed mood in this paper. We examine the influence of the intervention on the maternal patterns of depressed mood. We have previously documented that depressed intervention mothers breastfed their children longer, infants were taller and heavier at 6 months, and children were less likely to have Bayley Scales Scores reflecting cognitive disabilities (< 85) at 18 months, compared to infants of depressed mothers in the control condition (le Roux et al., 2013; Tomlinson et al., 2016). Thus, this article examines whether there are differences in the patterns of maternal depressed mood over time and whether the impact of these patterns on children varied, based on whether the mother received home visits or not.¹

Hypotheses

While we hypothesize that children of depressed mothers are more likely to demonstrate deficits, it is not clear from the existing literature whether maternal antenatal or postnatal depression results in greater problems. Based on the literature review above, we hypothesize that:

1. Depressed mothers have less income, education, are less frequently married or do not have a partner, and are more likely to experience intimate partner violence significantly more often than their non-depressed peers.

2. Children of mothers with depressed mood are shorter, weigh less, and are more likely to be stunted or malnourished than the peers whose mothers do not report depressed mood.

3. Children of mothers reporting both antenatal and postnatal depressed mood have significantly slower physical growth, slower language development, lower scores on executive functioning, and more behaviour problems than peers whose mothers are not depressed or only depressed at one assessment.

4. Children whose mothers are depressed more proximal to the assessment point show more significant deficits than children whose mothers only experience depressed mood antenatally.

Methods

All procedures were implemented following the guidelines and approved by the Institutional Review Boards of the University of California, Los Angeles, Stellenbosch University, and Emory University. All project employees were trained in ethics, with annual booster trainings.

Setting

In three peri-urban townships outside of Cape Town, South Africa, we matched 24 neighborhoods in blocks of four neighborhoods each, based on their size (450-600 households), the number of shebeens (typically 5-6), being less than 5 kilometers from a clinic, having a substantial distance or a natural barrier which ensured little contamination across neighborhoods, and to the presence (or not) of electricity, water, and toilets on the households’ premises. From May 2009 to September 2010, 98% of the pregnant women in these neighborhoods were recruited with voluntary informed consent and randomized by neighborhood to: 1) a control condition with standard clinic care (12 neighborhoods, 594 women); or 2) a Mentor Mother home-visiting intervention (12 neighborhoods, 644 women). Mothers were recruited at a mean of 26^th week of pregnancy (range (week 3-40). Over the first three years post-birth, 18 mothers and 100 of the children died; all were unrelated to study enrollment. By three years, more than 25% of the sample had migrated to the Eastern Cape, often with mothers leaving their children in the Eastern Cape to be raised by extended family. These mothers and children were re-assessed even if in the Eastern Cape. For the entire cohort, follow-up assessments were conducted at two weeks post-birth by 93% of mothers (mean=1.9 weeks; standard deviation (SD)=2.1 weeks); 87% at six months (mean=6.2 months; SD=0.7 months); 91% at 18 months (mean=19.1 months; SD=3.0 months); and 85% at 36 months (mean=37.3 months; SD=2.2 months). All assessments were completed by 80% of mothers; 1.6% (n=20) completed no follow-up re-assessments. Mothers and children could return to the study even if they missed an assessment. Follow-up rates were similar across conditions. Our analysis excluded mothers or children who died; most died within the first six months of birth and these participants are likely characteristically different. The flow of participants over time in the randomized control trial is available from the authors upon request.

Standard Care (SC) condition

Standard clinic care in Cape Town is accessible within 5 kilometers of each study neighborhood. Each antenatal clinic provides comprehensive maternal and child health services to prevent mother to child transmission of HIV. At the time this study was implemented, HIV clinics provided access to antiretroviral drugs for persons whose disease had compromised their immune system substantially (i.e., a cluster of differentiation count < 350).

Philani intervention condition

In addition to access to the clinic services available in the standard clinic condition, women in the intervention condition received home visits by community health workers (called Mentor Mothers). Township women who typically have never worked outside the home and who have less than a high school education were trained as Mentor Mothers for one month in cognitive-behavioral change strategies. Mentor Mothers were women selected to have good social and problem-solving skills, were mothers, and were trained to provide and apply health information about general maternal and child health, HIV, alcohol use, and nutrition. Videotaped models of common situations that Mentor Mothers might face were provided. Mentor Mothers were randomly observed during their home visits, received group supervision biweekly, and a monthly in-service training. Eight health messages were delivered on HIV/tuberculosis prevention, strategies for preventing mother to child HIV transmission, the consequences of alcohol use/abuse, the importance of breastfeeding, and how to avoid malnutrition. Mentor Mothers were to deliver these messages in at least four antenatal visits and four postnatal visits within the first two months of life. On average, Mentor Mothers made six antenatal visits (SD=3.8), five postnatal visits between birth and two months post-birth (SD=1.9), and about 1.4 visits/month (range: 0.1-6.4) until six months post-birth. Sessions lasted on average 31 minutes each. Visits were biannual after six months, primarily to check for growth and/or abuse.

Assessments

We recruited, trained, and certified township women as interviewers, entering responses on mobile phones (Nokia E61i and 2630) programmed by Mobenzi (http://www.mobenzi.com/researcher/). At each assessment, we monitored the following measures:

Maternal measures

Depressed mood

Depressed mood was reported at each assessment using the Edinburgh Postnatal Depression Scale (EPDS) (Cox, Holden, & Sagovsky, 1987). The EPDS is a 10-question self-rating scale that is administered to women as a means by which to screen for perinatal depression. Each question, which is scored on a continuum of 0-3 (where higher scores indicates greater distress), assesses one of the common characteristics of perinatal depression, as experiences in the last week. The final score, of between 0-30, indicates the level of distress associated with depression. It has been used across numerous cultural contexts (Husain et al., 2014; Patel, DeSouza, & Rodrigues, 2003b; Shrestha, Pradhan, Tran, Gualano, & Fisher, 2016) including in South Africa (Rochat, Tomlinson, Newell, & Stein, 2013; Tomlinson et al., 2014). In South Africa, the EPDS has demonstrated good internal reliability, with a Cronbach’s alpha of 0.87 (Hartley et al., 2011) and construct validity (De Bruin, Swartz, Tomlinson, Cooper, & Molteno, 2004).

Categories of depressed mood were created based on longitudinal profiles of depressed mood. Each mother was classified as never having a depressed mood (EPDS < 13 at all assessments); antenatally depressed only (EPDS > 13 only at the baseline assessment in pregnancy); postnatally depressed only (EPDS > 13 only post-birth); or depressed at both the antenatal assessment and at least one postnatal assessment (EPDS > 13 at the baseline and one additional assessment.

Demographic characteristics

Demographic characteristics measured at baseline were age, education, employment status, married or living with a partner, household income, whether experienced intimate partner violence (i.e., slapped, pushed or shoved, and/or threatened with a weapon by current partner), formal housing (vs. informal shack), water on the premises, a flush toilet, electricity, and food insecurity in the past week for mothers and separately rated by mothers for their children. Food insecurity was rated at recruitment as the number of days in the last week that you “didn’t have enough to eat” and then rated separately for their children. At the 36 month assessment only, we conducted a nine item measure of food insecurity on the Household Food Insecurity Access Scale (Swindale & Bilinsky, 2006). We found good construct validity with the single item measure for this sample (Tsai, Tomlinson, Comulada, & Rotheram-Borus, 2016), similar to researchers in other African countries (Tsai et al., 2012).

Health

Mothers reported at baseline whether they had previous pregnancies/childbirths and whether they had a previous low birth weight infant.

HIV status

HIV status was self-reported by mothers at each assessment; maternal status is also reflected on their child’s government-issued Road-to-Health card, as a check. Mothers could report a new HIV infection at a follow-up assessment during the study (i.e., HIV status could vary within mother). However, once identified as positive, mother’s HIV status was not allowed to change at later assessments. At recruitment, 325 of women were mothers living with HIV; over the next three years 22 additional mothers (2%) reported testing HIV seropositive.

Alcohol use

The frequency of drinking alcohol as well as the quantity was reported at each assessment. The Derived Alcohol Use Disorder Identification Test-Consumption (Derived AUDIT-C) (Saunders, Aasland, Babor, de la Fuente, & Grant, 1993), a three-item scale, each rated 0-4 on intensity, was completed which documents the severity of alcohol dependence when the score is greater than two (D. A. Dawson, Grant, & Stinson, 2005). Alcohol use prior to pregnancy discovery and alcohol dependence severity prior to and after pregnancy discovery was assessed separately.

Children’s adjustment

For each developmental measure, a psychologist with a strong developmental background trained and certified each interviewer as being a competent assessor. On-site supervision was provided weekly, including observations of the ongoing assessments.

Growth measures

At the post-birth, six month, 18 month, and 36 month assessments, trained and certified interviewers weighed children on scales recalibrated weekly and measured height and head circumference. Height, weight, and head circumference measures were then converted to z-scores based on the World Health Organization’s (WHO) age-adjusted norms (World Health Organization, 2006). Growth is reflected as standardized scores for height-for-age z-scores, weight-for-age z-scores, weight-for-length/height z-scores, and head circumference-for-age z-scores. A z-score less than −2 SD was considered a serious health deficit (de Onis & Blossner, 2003), as being stunted (< −2 for height-for-age) or malnourished (< −2 for weight-forage). About 6.5% (n=244) of children had height-for-age > 3 or < −3; 2.4% (n=92) of children had weight-for-age > 3 or < −3; similarly, 7.1% (n=265) had weight-for-length/height > 3 or < −3; and 4.2% (n=158) had head circumference-for-age > 3 or < −3. Growth scores values > 3 or < −3 may represent measurement errors and, therefore, these scores were omitted from the final analyses.

Child Behavior

At 36 months, mothers completed Achenbach’s Preschool Child Behavior Checklist (Achenbach, 1992), a 99-item questionnaire which yields problem scale scores for dimensions of internalizing and externalizing behaviors and also a total problems summary score. Children were considered deviant if their score was ≥ 14 for internalizing problems, ≥ 21 for externalizing problems, and ≥ 52 for the total number of behaviour problems. The Strengths and Difficulties Questionnaire (R. Goodman, 1997), also completed at 36 months, is a 25-item maternal report of behavioral adjustment, yielding two scale scores describing children’s prosocial behaviors and total difficulties. Children were prosocial behavior deviant if they scored ≤ 4 and were deviant on total difficulties if they scored ≥ 20.

The Peabody Picture Vocabulary Test

The Peabody Picture Vocabulary Test (Dunn, 1965); was administered at 36 months, using a form adapted for South Africa (Pakendorf & Alant, 1996).

Executive functioning

An executive functioning measure developed by Blair and colleagues was administered (Blair, Zelazo, & Greenberg, 2005) and was intended to reflect memory, inhibition, and attention shifting (Stone, Blumberg, Blair, & Cancelli, 2016). The Executive Function Battery was administered at 36 months and consists of three tasks: 1) Silly Sounds includes 36 questions where children were asked to identify pictures (e.g., a cat) that correspond with a sound (e.g., meow); 2) Something’s the Same includes 28 questions and children were asked to connect pictures based on common traits (e.g., color, size); and 3) Operation Span includes 16 questions and tested working memory. Final scores for each task were created by summing correct responses. This assessment measure has not been validated for South Africa; all segments are non-verbal, however.

Data analyses

Our analyses compared mothers and children over time, based on the reported patterns of maternal depressed mood. The never depressed category was treated as the reference group. We examined the distribution of each of the maternal characteristics at recruitment by depression category. Differences in depression categories were assessed using one-way ANOVA for continuous variables and chi-square tests for discrete variables. Pairwise comparisons were performed when an overall difference was observed between depression categories (i.e., at least one depression category was different from another). Bonferroni’s method for computing p-values and confidence intervals was used to account for multiple comparisons. Effect size (ES) estimates were provided for categories of depressed mood that were found to differ significantly from each other.

Maternal depressed mood was a key variable in this paper. Therefore, when a follow-up assessment was missing or if the EPDS score was missing within an assessment, then scores were imputed using multiple imputation. EPDS score at the baseline assessment was missing for 8.3%; follow-up assessments showed 8.2%, 8.0%, 9.7%, and 15.2% missing EPDS score at post-birth, 6 months, 18 months, and 36 months, respectively.

Missing scores for depressed mood were imputed under the assumption of missing at random (Rubin, 1987). Imputations were performed separately by intervention condition using multivariate normal regression which uses an iterative Markov chain Monte Carlo method. Compatible with the analysis model, we additionally controlled for maternal HIV status, alcohol use, food insecurity, and baseline education in the imputation model. We only included covariates which provided a unique source of variance as many markers of poverty were highly inter-correlated. Imputation was repeated 20 times and regression models were estimated separately using each imputed dataset. Results were then pooled according to the combination rules by Rubin (Rubin, 1987).

Longitudinal models were used to compare each measure of child growth assessed multiple times across depression categories and we accounted for repeated subject measures within neighborhoods (i.e., nested design nature) by including random intercepts for neighbourhood, as well as for child participant. Mixed-effects linear regression models with restricted maximum likelihood estimation (REML) were used to analyze continuous outcomes. Mixed-effects logistic regression models were used to analyze discrete outcomes. For behaviors where model convergence was not achieved, we removed the neighborhood random effect and instead included neighborhood as a fixed effect in the model. For measures only administered once (e.g., measures of child behavior and executive functioning), cross-sectional models were estimated to compare children’s outcomes based on their mother’s pattern of depression and included neighborhood random effects.

Longitudinal and cross-sectional models included fixed effects for the pattern of depressed mood and controlled for mother’s HIV status, alcohol use, food insecurity, and baseline education. Interactions between depression pattern and the control variables were also included. Child outcome measures were also adjusted for children’s age in months. Both longitudinal and cross-sectional models compared the four patterns of maternal depressed mood (never depressed, antenatal depression only, postnatal depression only, and antenatal/postnatal depression). Based on the pooled modeling results and for each pattern of depression, we predicted mean estimates for continuous measures and probability estimates for discrete outcome measures. Differences in depression categories were assessed using F-tests. Pairwise comparisons were performed when an overall mean difference was observed between depression categories (i.e., at least one depression category was different from another). Bonferroni’s corrections were used to account for multiple comparisons and effect size estimates were provided for depression patterns that were found to differ significantly from each other.

Twin and triplet births were excluded from the analyses as they were not comparable to singleton births (n=13). HIV+ children were also omitted (n=12). We used SAS 9.4 and Stata® SE software version 14.2 (StataCorp, 2015). Multiple imputation and regression analyses were performed using the -mi impute mvn-, -mi estimate-, -mixed-, and -melogit- commands in Stata.

Results

Patterns of depressed mood

The following depression patterns were identified: never reporting depressed mood (n=473; 42.3%); antenatal depressed mood only (n=154; 13.8%); postnatal depressed mood only (n=272; 24.3%); and antenatal and postnatal depressed mood (n=220; 19.7%). Only 9 mothers (0.8%) were depressed at each assessment. Among those with both antenatal and postnatal depression, the onset of postnatal depression was almost all in the first 6 months post-birth. Among those with both antenatal and postnatal depression (N=220), 63% (n=139) of the mothers were depressed within two weeks post-birth and 20% (n=46) were first depressed at 6 months. Only 35 mothers who were antenatally depressed, 9.7% (n=20) were postnatally depressed at 18 months and 6.8% (n=15) at 36 months post-birth. Examining when the onset of depression occurred for those depressed postnatally only (N=272), the first occasion of depression was post-birth for 32% (n=88), at 6 months for 27% (n=75), 20% (n=56) at 18 months, and 19% (n=53) at 36 months post-birth.

Table 1 summarizes the differences across patterns of depressed mood at the baseline assessment. Comparing the subgroups generally demonstrates that the mothers who never reported depressed mood were significantly better-off than their peers. Maternal age was significantly different across depression categories overall (p = 0.01) with antenatally depressed mothers younger than postnatally depressed mothers (p = 0.01; ES = 0.32). Education levels differed significantly overall (p < 0.01) with never depressed mothers having higher education than postnatally depressed mothers (p < 0.01; ES = 0.34) and mothers depressed both antenatally/postnatally (p < 0.01; ES = 0.50). Antenatally depressed mothers also had higher education than antenatally/postnatally depressed mothers (p = 0.01; ES = 0.32). Mothers differed by employment overall (p = 0.03) with those never depressed having higher employment compared to antenatally/postnatally depressed mothers (p = 0.01; ES = −0.09). Household income was significantly different by depressed mood overall (p < 0.01). Significantly more never depressed mothers had incomes > 2000 rand per month compared to antenatally depressed mothers (p = 0.04; ES = −0.13) and antenatally/postnatally depressed mothers (p < 0.01; ES = −0.20). Postnatally depressed mothers also had higher incomes compared to antenatally/postnatally depressed mothers (p < 0.01; ES = −0.16). Mothers who experienced intimate partner violence differed overall across depression categories (p < 0.01). Never depressed mothers were less likely to experience intimate partner violence compared to antenatally/postnatally depressed mothers (p < 0.01; ES = 0.21). Postnatally depressed mothers were also less likely to experience intimate partner violence compared to antenatally/postnatally depressed mothers (p = 0.04; ES = 0.12). Mothers differed overall by whether they had electricity on site (p < 0.01) with never depressed mothers more likely to have electricity compared to antenatally/postnatally depressed mothers (p = 0.01; ES = −0.09). Maternal food insecurity differed significantly overall across depression categories (p < 0.01). Never depressed mothers were less likely to experience food insecurity than antenatally depressed (p < 0.01; ES = 0.16), postnatally depressed (p < 0.01; ES = 0.14), and antenatally/postnatally depressed (p < 0.01; ES = 0.24) mothers. Likewise, there was a significant difference overall in children of pregnant women experiencing food insecurity across depression categories (p < 0.01). Never depressed mothers were less likely to experience food insecurity than antenatally depressed (p < 0.05; ES = 0.11), postnatally depressed (p < 0.01; ES = 0.13), and antenatally/postnatally depressed (p < 0.01; ES = 0.21) mothers. Never depressed mothers were significantly less likely to have had previous children compared to antenatally/postnatally depressed mothers (p = 0.04; ES = 0.10). Antenatally depressed mothers were also less likely to have other children compared to postnatally depressed mothers (p = 0.03; ES = −0.14) and antenatally/postnatally depressed mothers (p = 0.01; ES = 0.16).

Table 1.

Baseline characteristics of the sample (N=1,119) summarized by pattern of depression

	Never (N=473)		Antenatal (N=154)		Postnatal (N=272)		Ante/Postnatal (N=220)		Total (N=1,119)
	n	%	n	%	n	%	n	%	n	%
Demographic characteristics
Age, mean (SD)*	26.2	5.5	25.33	5.2	27.22	6.0	26.6	5.3	26.4	5.6
Highest education level, mean (SD)*	10.73,4	1.7	10.44	1.7	10.11	1.7	9.81,2	2.1	10.4	1.8
Currently employed*	1094	23.0	32	20.8	49	18.0	301	13.6	220	19.7
Married or lives with partner	276	58.4	76	49.4	168	61.8	120	54.5	640	57.2
Monthly household income > 2000 rand*	2522,4	54.8	611	42.1	1354	50.6	721,3	34.3	520	48.1
Formal housing	153	32.3	57	37.0	85	31.3	54	24.5	349	31.2
Water on site	257	54.3	89	57.8	144	52.9	105	47.7	595	53.2
Flush toilet on site	264	55.8	91	59.1	154	56.6	107	48.6	616	55.0
Electricity on site*	4384	92.6	131	85.1	246	90.4	1841	83.6	999	89.3
Mother hungry past week*	1842,3,4	38.9	841	54.5	1451	53.3	1391	63.2	552	49.3
Children hungry past week*	952,3,4	20.1	481	31.2	891	32.7	911	41.4	323	28.9
Maternal health
Non-primipara*	2964	62.6	883,4	57.1	1932	71.0	1601,2	72.7	737	65.9
Previous low birth weight infant	43	14.5	17	19.3	28	14.5	29	18.1	117	15.9
Intervention conditions
Philani intervention*	229	48.4	92	59.7	133	48.9	128	58.2	582	52.0
Standard care	244	51.6	62	40.3	139	51.1	92	41.8	537	48.0
Partnerships and HIV
Recent sexual partnera	425	89.9	134	87.0	245	90.1	192	87.3	996	89.0
Ever tested for HIV	444	93.9	148	96.1	259	95.2	201	91.4	1,052	94.0
Mothers living with HIV*	1083,4	24.5	46	31.1	911	35.1	801	40.0	325	31.0
Recent sexual partnera	94	87.0	37	80.4	79	86.8	70	87.5	280	86.2
Disclosed HIV status to partner	58	77.3	27	81.8	52	78.8	41	74.5	178	77.7
Number of people disclosed to, mean (SD)	4.4	5.9	3.6	3.0	4.6	6.3	4.0	4.3	4.2	5.3
Alcohol
Drank any alcohol in the month prior to pregnancy discovery*	93	22.9	40	26.0	484	19.6	693	31.4	250	24.4
AUDIT-C > 2, month prior to pregnancy discovery*	66	16.2	29	18.8	364	14.7	553	25.0	186	18.1
AUDIT-C > 2, after pregnancy discovery*	18	4.4	12	7.8	10	4.1	22	10.0	62	6.0

SD: standard deviation; HIV: human immunodeficiency virus; AUDIT-C: Alcohol Use Disorders Identification Test-Consumption

^a“Recent” refers to the last three months

^*p < 0.05 (one-way ANOVA for continuous variable or χ² tests for discrete variables)

Superscript numbers indicate pairwise comparison significant differences (p < 0.05) after accounting for multiple comparisons using Bonferroni’s method for adjustment:

¹Never;

²Antenatal;

³Postnatal;

⁴Ante/Postnatal

Being HIV seropositive was associated with depression (p < 0.01); never depressed mothers were less likely to be living with HIV compared to postnatally depressed (p = 0.02; ES = 0.11) and antenatally depressed mothers (p < 0.01; ES = 0.16). Depressed mothers differed significantly overall in alcohol use (p = 0.02) with postnatally depressed mothers less likely to use any alcohol compared to antenatally/postnatally depressed mothers (p = 0.02; ES = 0.12). Problematic alcohol use (AUDIT-C > 2) prior to pregnancy discovery varied significantly based on depression (p = 0.02). Postnatally depressed mothers were less likely to have problematic drinking compared to antenatally/postnatally depressed mothers (p = 0.03; ES = 0.10). Finally, problematic alcohol use after discovering one is pregnant was significantly different based on maternal depression (p = 0.02). However, pairwise comparisons revealed no differences after multiple comparisons adjustment using Bonferroni’s method.

There were no significant differences in the patterns of depressed mood based on intervention condition.

Children’s adjustment

Table 2 summarizes the differences in children’s adjustment across maternal category of depressed mood. The measures are summarized based on the results from 20 imputations (i.e., 20 different imputed datasets). The estimates provided are pooled over results from separate models run on each imputed dataset with the frequency of the depression category varying between each imputation. Thus, we are not able to provide a sample size distribution for maternal depression categories due to the variation that occurs in each imputation.

Table 2.

Child outcomes summarized by pattern of depression

	Never		Antenatal		Postnatal		Ante/Postnatal		Total
	Mean	SE	Mean	SE	Mean	SE	Mean	SE	Mean	SE	P-valuea
Language
Peabody Picture Vocabulary Test score	19.79	0.53	19.66	0.79	19.20	0.59	19.89	0.73	19.68	0.32	0.85
Growth measures
Height-for-age z-score	−0.45	0.04	−0.41	0.07	−0.49	0.05	−0.62	0.06	−0.49	0.03	< 0.05
Weight-for-age z-score	0.324	0.05	0.23	0.07	0.14	0.05	0.071	0.06	0.21	0.03	0.01
Weight-for-length/height z-score	0.71	0.04	0.72	0.07	0.60	0.05	0.59	0.06	0.66	0.02	0.19
Head circumference-for-age z-score	0.43	0.04	0.34	0.07	0.38	0.05	0.24	0.06	0.37	0.02	0.06
Behavior
Child Behavior Checklist
Internalizing Problems Scale score	11.993,4	0.48	13.87	0.74	14.671	0.56	15.641	0.68	13.73	0.28	< 0.01
Externalizing Problems Scale score	16.204	0.53	18.42	0.79	17.65	0.61	19.401	0.74	17.58	0.32	< 0.01
Total Problems score	42.932,3,4	1.32	49.961	2.01	49.921	1.55	53.501	1.90	48.08	0.75	< 0.01
Strengths and Difficulties Questionnaire
Prosocial Behavior Subscale score	7.49	0.14	7.60	0.21	7.09	0.16	7.29	0.20	7.39	0.10	0.14
Total Difficulties score	8.382,3,4	0.28	9.871	0.41	10.021	0.31	11.091	0.38	9.61	0.16	< 0.01
Executive function
Executive Function Battery
Silly Sounds score	5.91	0.45	6.11	0.68	5.47	0.52	5.32	0.62	5.70	0.25	0.77
Something’s the Same score	4.30	0.42	5.21	0.64	4.29	0.49	4.17	0.60	4.38	0.24	0.62
Operation Span score	1.44	0.15	1.27	0.24	0.984	0.18	1.923	0.21	1.34	0.09	0.01
	Prob.	SE	Prob.	SE	Prob.	SE	Prob.	SE	Prob.	SE	P-valueb
Growth measures
Height-for-age z-score < −2 SD	0.09	0.01	0.10	0.02	0.09	0.01	0.13	0.02	0.10	0.01	0.06
Weight-for-age z-score < −2 SD	0.02	0.00	0.01	0.00	0.03	0.01	0.02	0.01	0.02	0.00	0.07
Behavior
Child Behavior Checklist
Internalizing Problems Scale score deviant	0.343,4	0.03	0.42	0.05	0.491	0.04	0.571	0.04	0.44	0.02	< 0.01
Externalizing Problems Scale score deviant	0.274	0.03	0.37	0.05	0.34	0.03	0.411	0.04	0.33	0.02	0.02
Total Problems score deviant	0.273,4	0.03	0.38	0.05	0.411	0.04	0.441	0.04	0.36	0.02	< 0.01
Strengths and Difficulties Questionnaire
Prosocial Behavior Subscale score deviant	0.10	0.02	0.09	0.03	0.15	0.03	0.13	0.03	0.12	0.01	0.40
Total Difficulties score deviant	0.014	0.01	0.024	0.01	0.06	0.01	0.091,2	0.02	0.04	0.01	0.04

SE: standard error; SD: standard deviation

^aTest of whether depression category mean estimates produced from mixed-effects linear regression models for continuous outcomes are jointly equal

^bTest of whether depression category probability estimates produced from mixed-effects logistic regression models for binary outcomes are jointly equal

Superscript numbers indicate pairwise comparison significant differences (p < 0.05) after accounting for multiple comparisons using Bonferroni’s method for adjustment:

¹Never;

²Antenatal;

³Postnatal;

⁴Ante/Postnatal

Children of mothers with different depression patterns varied significantly on the height-for-age (F = 2.6, p < 0.05) and weight-for-age (F = 4.1, p = 0.01) z-scores. Pairwise comparisons of patterns of depression for height-for-age z-scores, however, revealed no differences after multiple comparisons adjustment using Bonferroni’s method. Children of never depressed mothers had higher weight-for-age z-scores than children of antenatally/postnatally depressed mothers (p = 0.01; ES = −0.08). Weight-for-length/height and head circumference-for-age were not significantly different overall across patterns of maternal depression. The percentage of stunted (< −2 SD) and malnourished children (< −2 SD) was similar across different patterns of depression.

Figures 1 and 2 plot the differences in growth scores for height-for-age and weight-for-length/height across mothers with different patterns of depression based on the results using the first imputed dataset. Each of the growth measures were significantly different on height-for-age (F = 5.6, p < 0.01), weight-for-age (F = 15.2, p < 0.01), weight-for-length/height (F = 3.2, p = 0.02), and head circumference-for-age (F = 5.3, p < 0.01). Height-for-age and weight-for-age Z scores are plotted over time for the readers on Figure 1 and 2 respectively. These figures demonstrate the growth patterns over time, based on the timing of maternal depressed mood/or not. In addition to demonstrating growth differences based on the pattern of maternal depression, the graphs also demonstrate a community-wide drop in height-for-age by all children from the 18 to the 36 month assessment. Children of antenatally/postnatally depressed mothers have a lower height-for-age and weight-for-length/height z-scores compared to every other subgroup at which mothers were depressed at least at one assessment point.

Figure 1.

Plots of the height-for-age Z scores at each assessment point (Pregnancy, birth, 6 months, 18 months, and 36 months post-birth) based on the pattern of maternal depression (antenatal depressed mood only, postnatal depressed mood only, antenatal and postnatal depressed mood, and no depressed mood at any assessment).

Figure 2.

Plots of the weight-for-age Z scores at each assessment point (Pregnancy, birth, 6 months, 18 months, and 36 months post-birth) based on the pattern of maternal depression (antenatal depressed mood only, postnatal depressed mood only, antenatal and postnatal depressed mood, and no depressed mood at any assessment).

There was no significant difference on the Peabody Picture Vocabulary Test at 36 months. There were significant differences on each measure of child behavioral adjustment. Significant differences were observed overall in the Child Behavior Checklist Internalizing Problems Scale Score (F = 7.8, p < 0.01), Externalizing Problems Scale Score (F = 7.8, p < 0.01), and Total Problems Score (F = 8.4, p < 0.01). On the Internalizing measure, children of never depressed mothers had lower scores compared to children of postnatally depressed mothers (p < 0.01; ES = 0.18) and antenatally/postnatally depressed mothers (p < 0.01; ES = 0.21). On the Externalizing measure, children of never depressed mothers had lower scores compared to children of antenatally/postnatally depressed mothers (p < 0.01; ES = 0.17). For the Total Problems Score, there were significantly fewer symptoms for children of mothers who never had depressed mood compared to children of antenatally depressed (p = 0.02; ES = 0.14), postnatally depressed (p < 0.01; ES = 0.17), and antenatally/postnatally depressed (p < 0.01; ES = 0.22) mothers. There were also significant differences overall on Internalizing symptoms (F = 7.2, p < 0.01), the Externalizing symptoms (F = 3.2, p = 0.02), and the Total Problems symptoms (F = 5.6, p < 0.01). On the Internalizing deviant measure, children of never depressed mothers had lower probability of symptoms compared to children of postnatally depressed mothers (p < 0.01; ES = 0.15) and antenatally/postnatally depressed mothers (p < 0.01; ES = 0.23). On the Externalizing deviant measure, children of never depressed mothers had lower probability compared to children of antenatally/postnatally depressed mothers (p = 0.03; ES = 0.14). There was also a lower probability of being deviant on the Total Problems Score for children of never depressed mothers compared to children of postnatally depressed mothers (p = 0.01; ES = 0.14) and antenatally/postnatally depressed mothers (p < 0.01; ES = 0.17).

Children varied significantly overall on the Total Difficulties Subscale of the Strengths and Difficulties Questionnaire (F = 12.6, p < 0.01) across subgroups of maternal depression. Never depressed mothers had children who had fewer problems than children of antenatally depressed (p = 0.02; ES = 0.15), postnatally depressed (p < 0.01; ES = 0.19), and antenatally/postnatally depressed (p < 0.01; ES = 0.28) mothers. On the Total Difficulties deviant measure, there was a significant difference overall (F = 2.8, p = 0.04) with children of never depressed mothers having lower probability of being deviant compared to children of antenatally/postnatally depressed mothers (p = 0.01; ES = 0.08). Children of antenatally depressed mothers also had lower probability of being deviant compared to children of antenatally/postnatally depressed mothers (p < 0.05; ES = 0.07). However, there were not significant differences overall across maternal depression categories in children’s mean Prosocial Behavior Subscale Score and in probabilities of children being disordered.

Executive functioning as measured by the Executive Function Battery at 36 months did not show any significant differences overall across maternal depressed mood based on the Silly Sounds and Something’s the Same tasks, but there were significant differences overall on the Operation Span Score (F = 4.0, p = 0.01). Children of postnatally depressed mothers scored lower on the Operation Span task compared to children of antenatally/postnatally depressed mothers (p < 0.01; ES = 0.17).

We examined 15 measures based on intervention condition and depressed mood. For intervention main effects, there was only one significant difference and when examining the intervention by depression condition interaction only two differences approached significance. The probability that this significant difference could have occurred by chance is substantial, therefore, we are not reporting the single intervention effect.

Discussion

About 250 million children under 5 years in high income countries whose mothers are depressed will not meet their developmental potential (M. M. Black et al., 2017). Our study provides powerful confirmatory evidence that the roots of these challenges are in early life, from pregnancy through at least the first three years of life. Most of the mothers in this population sample (58%) experienced depressed mood either while pregnant or at some point over the first three years of their child’s life. While only one in five mothers experiences both antenatal and postnatal depressed mood during the first three years post-birth, there is an additional 14% experiencing only antenatal depressed mood and 24% experiencing depressed mood after childbirth within the next three years. These data suggest the importance of community-wide interventions for depression for mothers – rather than focussing on a specific time point such as antenatally or in the early postnatal period. Regardless of when maternal depression occurs - antenatally, postnatally, or both - the children of depressed mothers have developmentally-linked challenges.

Less than 1% of mothers reported depression at each assessment, suggesting that “chronic” depression may not be a useful label for depressed mothers during the perinatal period. The initial research that established this label for mothers depressed both antenatally and postnatally (Mora et al., 2009; Sutter-Dallay et al., 2012) did not have as many assessments of maternal depression as was obtained in this study.

Depressed mood can be either a cause or a consequence of more structural or exogenous disadvantages. In this study, mothers with depressed mood experienced significantly more stressors than mothers without depressed mood. Mothers without depressed mood were more educated, were likely to have previous children, a partner living at home and to have electricity. Food insecurity, multiple children, alcohol use/abuse, HIV status, and intimate partner violence emerge as important factors associated with maternal depression. Addressing these stressors among pregnant and new mothers may significantly improve children’s adjustment when their mothers are depressed.

The Philani home visiting intervention did not significantly affect the patterns of mothers’ reports of depressed mood antenatally or through the first 36 months of life. Earlier we have shown that depressed, intervention mothers living with HIV were 50% more likely to complete the tasks to prevent transmission of HIV, their children had better growth in the first six months of life, and depressed mothers breastfed significantly longer than the depressed mothers in the control condition (le Roux et al., 2013; Tomlinson, Rotheram-Borus, Harwood, & le Roux, in press). During the period that the Mentor Mothers were routinely visiting the households (pregnancy and the first six months of life), the children of antenatally depressed mothers receiving home visits had higher height-for-age and weight-for-age z-scores than children of antenatally depressed mothers in the control condition (Tomlinson et al., 2015). However, this advantage was lost prior by 18 months post birth.

We did find significant differences in behavior problems across depression subgroups on both the Achenbach Child Behavior Checklist and the Strengths and Difficulties Questionnaire. Mothers with different patterns of depressed mood had children with different levels of both internalizing and externalizing symptoms and disorders. Similar to the conclusion of Rahman and colleagues (2007), more depressed mood appears to influence children’s adjustment negatively, at least in terms of child behavior problems. While children with behavior problems are challenging for mothers (which can lead to more maternal depressed mood), the children in this sample are actually too young to observe the serious deficits that children of depressed mothers are likely to occur when they enter school. Children with behavior problems are far less likely to achieve in school (Sayal, Washbrook, & Propper, 2015) and at three years post-birth, the children are already demonstrating behaviour problems. We are currently assessing this cohort as they enter school. These data are particularly important in demonstrating the negative behavioural consequences of having a depressed mother.

Surprisingly, there were no significant differences based on patterns of maternal depressed mood on the Peabody Picture Vocabulary Test or on two of three measures of Blair and colleagues (2005) of executive functioning. The Peabody Picture Vocabulary Test is well validated scale that has been used many times in South Africa (Campbell, Bell, & Keith, 2001; Murray, Cooper, Arteche, Stein, & Tomlinson, 2015). The executive functioning measure, with multiple variations on match-to-sample tasks, had skewed distributions and has not been used with international samples. It is not clear how well this measure may tap executive functioning among South African children at three years of age. The tasks aimed to be culturally free of bias (silly sounds, animals) and are non-verbal and not based on language ability. Yet, it is not clear when assessing concepts, which are evolving quickly in children and among children of a very different ethnic, racial, and socioeconomic background, whether the measure of executive functioning was valid.

This was a population cohort and there was both high internal and external validity among the sample participants over three years. Almost all mothers in a neighborhood participated (98%). There were only 20 mothers (1.6%) of the sample who were not reassessed at least once after recruitment; 80% were assessed across all time points. We did not conduct diagnostic interviews for depression and the EPDS cut-off score can only suggest that a mother is depressed. A major limitation of this study is the use of the EPDS. It is only a screening measure of depressed mood and therefore does not provide a diagnosis of depression, Furthermore, the time-span for which the symptoms are reported is only the last week. Mothers who did not meet the established criteria for depressed mood (score > 13) may have been depressed at other time points over the first three years of life, not covered by the EPDS as the last week prior to the reassessment. The categorization of depression may be similarly flawed because of the short time frame of the EPDS for depressed mood and the failure to conduct diagnostic interviews.

In summary, depressed mood among this population sample of township mothers in South Africa appears to have significant impacts on children’s growth and behavior. With depression so common in the community, addressing the environmental challenges faced by township mothers who become depressed could potentially reduce the effects of maternal depression. In particular, food insecurity, alcohol use/abuse, and HIV are areas for intervention, which may be as productive as cognitive-behavioral interventions for depression.When significant differences were found, children of never depressed mothers do better than their peers and children whose mothers are depressed both antenatally/postnatally do the worst. More proximal depressed mood appears to have a greater impact on children. Unfortunately, the intervention did not shift patterns of depressed mood over time nor did it impact child adjustment, related to the pattern of maternal depressed mood. Maternal depressed mood, experienced by more than half of township mothers in the first three years of their children’s life remains a significant challenge. Structural interventions are likely to be needed, not only community-level depression treatment.

Public Health Significance.

The data from this study demonstrate that economic resources and maternal mental health problems, HIV and problematic alcohol use are associated with patterns of maternal depression over time, as well as deficits in child growth and behaviour.