Introduction
Maternal depressive symptoms, whether meeting the diagnostic criteria for major depressive disorder (MDD) or not, are highly prevalent, with estimates ranging from 3% to 60% (Walker et al. Reference Walker, Wachs, Grantham-Mcgregor, Black, Nelson, Huffman, Baker-Henningham, Chang, Hamadani, Lozoff, Gardner, Powell, Rahman and Richter2011) across a variety of assessment tools, timing of screening, and study populations worldwide (Marcus, Reference Marcus2009). In a US population, screening of pregnant women demonstrated that over 20% have elevated depressive symptomatology (Marcus et al. Reference Marcus, Flynn, Blow and Barry2003). A history of depressive symptoms, poor overall health, high alcohol use, smoking, single marital status, unemployment, and low educational attainment have been significantly associated with depressive symptoms during pregnancy (Marcus et al. Reference Marcus, Flynn, Blow and Barry2003). When a woman has high depressive symptoms, it means that her score is above the cut-off point of the screening or diagnostic measure that is used to assess her depressive symptoms. Mothers with high depressive symptoms tend to be less sensitive and more negative when interacting with their infants than mothers with few or no depressive symptoms (Cooper et al. Reference Cooper, Tomlinson, Swartz, Woolgar, Murray and Molteno1999).
Children's cognitive development in early childhood is affected by genetic, biological, social, and psychological factors, many of which are sensitive to broader contextual variables such as poverty, cultural norms, and child-rearing environments (Grantham-McGregor et al. Reference Grantham-Mcgregor, Cheung, Cueto, Glewwe, Richter and Strupp2007; Walker et al. Reference Walker, Wachs, Grantham-Mcgregor, Black, Nelson, Huffman, Baker-Henningham, Chang, Hamadani, Lozoff, Gardner, Powell, Rahman and Richter2011). Since mothers largely influence infants’ social environments and mediate their experiences with the external world, maternal sensitivity, responsiveness and affect (emotional warmth or rejection) are consistently related to young children's cognitive and social-emotional competence (Walker et al. Reference Walker, Wachs, Meeks Gardner, Lozoff, Wasserman, Pollitt and Carter2007).
Although a meta-analysis of 46 observational studies found that maternal depressive symptoms increased the likelihood of negative parenting behaviors including impatience, low sensitivity, hostility and overall negative parent–child interactions, (Lovejoy et al. Reference Lovejoy, Graczyk, O'HARE and Neuman2000) findings have been mixed regarding relationships between maternal depressive symptoms and cognitive development in early childhood. Hay & Kumar (1995) reported an association between postnatal depression and impaired child cognitive abilities at 4 years of age in a North London community sample. Similarly, Cornish and colleagues showed that infants of chronically depressed mothers were 3.4 times more likely to receive a non-optimal score on the Mental Development Index (MDI) of the Bayley Scales of Infant and Toddler Development, Second Edition (BSID-II; Cornish et al. Reference Cornish, Mcmahon, Ungerer, Barnett, Kowalenko and Tennant2005). A recent study also found that children of mothers having had a depressive episode at 6 weeks postpartum had significantly lower BSID-II MDI scores at 18 months compared with children of mothers with no depressive episodes (Conroy et al. Reference Conroy, Pariante, Marks, Davies, Simone, Schacht and Moran2012). In contrast, Kurstjens & Wolke (2001) concluded that maternal depression had negligible effects on children's cognitive development at 4 years and 8 months, and Hanley et al. (2013) reported that there was no significant difference observed in any of the BSID-III subscales (p > 0.05) between infants exposed and unexposed to pre- and postnatal maternal depressed mood.
Although a few descriptive reviews have summarized the research findings linking maternal depressive symptoms to young children's cognitive development (Field, Reference Field1992; Grace et al. Reference Grace, Evindar and Stewart2003; Sohr-Preston & Scaramella, Reference Sohr-Preston and Scaramella2006; Wachs et al. Reference Wachs, Black and Engle2009), and a meta-analysis of nine studies in 1998 showed that postpartum depression had a small but significant effect on children's cognitive and emotional development at age of ⩾1 year (Beck, Reference Beck1998), to the best of our knowledge, no quantitative synthesis of recent research results has been produced examining this association in early childhood. The objectives of the present study were to systematically review the literature on maternal depressive symptoms and young children's cognitive development and to summarize the associations found across populations using meta-analytical techniques.
Method
This study follows the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA; Moher et al. Reference Moher, Liberati, Tetzlaff and Altman2009) and Meta-Analysis of Observational Studies in Epidemiology (MOOSE) statement for standard reporting. (Stroup et al. Reference Stroup, Berlin, Morton, Olkin, Williamson, Rennie, Moher, Becker, Sipe and Thacker2000).
Study inclusion criteria
A study was considered eligible for inclusion in this meta-analysis if it: (i) quantitatively assessed relationships between maternal depression or depressive symptoms and child cognitive development; (ii) was published in a peer-reviewed journal; (iii) was not a case study; and (iv) assessed the outcome in young children aged between 0 and 7 years, since our study was designed to capture cognitive developmental process in early childhood; and (v) had a sample that was representative of the general population of young children in that context. If the study population was described as a specific subset of children (e.g. obese children), the study was considered not representative of the general population and thus excluded. We applied no other population, language or time restrictions.
Search strategy
Studies included in this meta-analysis were identified using both electronic and manual searches. Electronic databases were searched for relevant studies on maternal depressive symptoms and child cognitive development: PubMed, EMBASE (without explosion) and PsycINFO. A string of search terms was applied to the three databases: (‘cognition’ or ‘cognitive’ or ‘psychomotor’ or ‘sensorimotor’ or ‘motor’ or ‘child development’ or ‘infant development’ or ‘intelligence’ or ‘IQ’ or ‘visual’ or ‘vision’ or ‘language’ or ‘executive function’ or ‘attention’ or ‘memory’ or ‘school readiness’ or ‘pre-academic’ or ‘academic’ or ‘mental’ or ‘brain’) AND (‘maternal depression’ OR ‘maternal depressive symptoms’). Although we were initially interested in studies capturing both motor and cognitive development, the relatively low numbers of studies examining motor processes led us to focus exclusively on cognition. Follow-up manual searches were conducted from the citations in the review and meta-analysis articles retrieved from the electronic search.
Study selection and data extraction
The selection of studies was conducted in three steps. First, titles and abstracts of the retrieved studies were screened independently by two reviewers (Y.L. and J.C.). Studies deemed by both reviewers to not fulfill the criteria were excluded. Second, the remaining studies were read in full by the two reviewers and selected for inclusion in the analysis by consensus. Authors were contacted if the full text of a potentially relevant article was not available. Articles identified as relevant by both reviewers were included. Disagreements between the two reviewers over study eligibility were resolved by a third reviewer (M.F.), who independently reviewed the articles with discordant assessments, discussed findings with the two reviewers, and participated in the group consensus. Reasons for exclusion were recorded and reviewed by the group. Third, data from each selected study were abstracted onto a standardized form independently by the two reviewers, including the authors, year of publication, the country where the study was conducted, study population, sample size, study design, the exposure and outcome measures, the age of the child at each assessment, effect estimate, confounders, and limitations. The summary results were compared and discussed, and any discrepancies were resolved by the third reviewer.
Meta-analysis
The meta-analysis was conducted using Stata v. 14 (StataCorp., USA). All studies reported child cognitive development as continuous variables, using a variety of measures. We used ‘maternal depressive symptoms’ as our exposure to capture studies using either a measure of depressive symptoms or a diagnostic measure of depression. To examine the crude association of maternal depressive symptoms and child cognitive development (i.e. the unadjusted analysis), we converted estimates of effect sizes to a common metric of Cohen's d (Cohen, Reference Cohen1988), which represents the difference between two group means (cognitive scores for children of mothers with high v. low/no depressive symptoms) divided by the pooled standard deviation (s.d.); stratified analyses by study type and outcome measure were also conducted to minimize the heterogeneity among studies. In addition, the studies that reported regression coefficients from multivariate linear regression as their effect size measures were also synthesized (i.e. the adjusted analysis), which enabled us to assess the relationship between maternal depressive symptoms and child cognitive development, adjusting for a set of possible confounders.
Since few studies assessed specific aspects of cognitive development (e.g. memory, attention), we focused on measures of ‘general’ cognitive development as our outcome of interest. If a study assessed depressive symptoms at several time points, we used the postnatal assessment with the longest follow-up time due to the fact that the data was correlated between time points and the availability of data. When a study used more than one measure to screen depressive symptoms, we gave priority to the measures that were more commonly used in the literature.
We used standard meta-analytical methods to estimate the summary effect sizes using the inverse variance approach and random-effects models. Heterogeneity was assessed by calculating the Q statistic and I 2 index, and by conducting subgroup analyses to determine if point estimates differed among groups. Publication bias was assessed from funnel plots using both the Egger's linear regression test and Begg's rank correlation test to determine statistical significance.
Results
The search terms retrieved 764 citations from PubMed, 1327 citations from EMBASE, and 1443 citations from PsycINFO. Additional relevant articles were found through manual search. We identified 95 potentially relevant studies by screening the titles and abstracts retrieved from the electronic and manual searches (Fig. 1). Among these, 14 studies meeting the predefined selection criteria were included in the crude analysis (Supplementary Table S1). Studies were conducted in ten countries: three in the UK (Hay & Kumar, Reference Hay and Kumar1995; Conroy et al. Reference Conroy, Pariante, Marks, Davies, Simone, Schacht and Moran2012; Husain et al. Reference Husain, Cruickshank, Tomenson, Khan and Rahman2012), two in the United States (Cicchetti et al. Reference Cicchetti, Rogosch, Toth and Spagnola1997; Pound, Reference Pound2005), two in Australia (Milgrom et al. Reference Milgrom, Westley and Gemmill2004; Cornish et al. Reference Cornish, Mcmahon, Ungerer, Barnett, Kowalenko and Tennant2005), and one each in Japan (Otake et al. Reference Otake, Nakajima, Uno, Kato, Sasaki, Yoshioka, Ikeno and Kishi2014), Canada (Hanley et al. Reference Hanley, Brain and Oberlander2013), France (Sutter-Dallay et al. Reference Sutter-Dallay, Murray, Dequae-Merchadou, Glatigny-Dallay, Bourgeois and Verdoux2011), Germany (Kurstjens & Wolke, Reference Kurstjens and Wolke2001), Greece (Koutra et al. Reference Koutra, Chatzi, Bagkeris, Vassilaki, Bitsios and Kogevinas2013), India (Patel et al. Reference Patel, Desouza and Rodrigues2003) and Bangladesh (Black et al. Reference Black, Baqui, Zaman, Mcnary, Le, Arifeen, Hamadani, Parveen, Yunus and Black2007) Two studies were cross-sectional, and the other 12 studies used longitudinal designs, among which eight studies assessed maternal depressive symptoms at least 4 months prior to the assessment of child cognitive development. All of these 14 studies provided enough data for us to examine the crude association between maternal depressive symptoms and young children's cognitive development, but this association could potentially be biased due to the unadjusted confounding. Among these studies, four studies that focused on the MDI from the BSID also reported unstandardized regression coefficients from multivariate linear regression, which allowed us to adjust for potential confounders. (Black et al. Reference Black, Baqui, Zaman, Mcnary, Le, Arifeen, Hamadani, Parveen, Yunus and Black2007; Sutter-Dallay et al. Reference Sutter-Dallay, Murray, Dequae-Merchadou, Glatigny-Dallay, Bourgeois and Verdoux2011; Conroy et al. Reference Conroy, Pariante, Marks, Davies, Simone, Schacht and Moran2012; Koutra et al. Reference Koutra, Chatzi, Bagkeris, Vassilaki, Bitsios and Kogevinas2013) There were no studies of children beyond 56 months of age that fulfilled the criteria to be included in the analysis. Although the majority of studies employed representative samples, one study (Kurstjens & Wolke, Reference Kurstjens and Wolke2001) included 1011 children (76.1%) with special health needs who were admitted to special care units within the first 10 days of life into their initial study sample of 1329 children. The removal of this study from the primary analysis did not change the results significantly [Cohen's d = −0.28, 95% confidence interval (CI) −0.42 to −0.13]. Two of the 14 studies (Black et al. Reference Black, Baqui, Zaman, Mcnary, Le, Arifeen, Hamadani, Parveen, Yunus and Black2007 and Sutter-Dallay et al. Reference Sutter-Dallay, Murray, Dequae-Merchadou, Glatigny-Dallay, Bourgeois and Verdoux2011) in the unadjusted analysis assessed cognitive development at multiple time points. Due to the fact that the data was correlated between time points and the availability of data, we chose the longest time point for the two studies (12 months for Black et al. Reference Black, Baqui, Zaman, Mcnary, Le, Arifeen, Hamadani, Parveen, Yunus and Black2007 and 24 months for Sutter-Dallay et al. Reference Sutter-Dallay, Murray, Dequae-Merchadou, Glatigny-Dallay, Bourgeois and Verdoux2011).
Flow diagram of study selection procedure. The reasons to exclude initially identified articles to reach at n = 95 were: (1) the study was non-relevant to our research questions; (2) the study was a narrative review or description for a cohort; (3) study participants were beyond age range or not representative of general population; (4) study outcomes did not focus on general cognitive development.
The majority of studies used the MDI from the BSID-II or BSID-III (Bayley, Reference Bayley1969, Reference Bayley1993, Reference Bayley2006a , Reference Bayley b ) to measure child cognitive development (n = 10). Other studies used the Developmental Assessment Scale for Indian Infants (DASII, n = 1) (Patel et al. Reference Patel, Desouza and Rodrigues2003), McCarthy Scales of Children's Abilities (n = 1) (McCarthy, Reference Mccarthy1972), the Early Screening Profiles (n = 1) (Harrison et al. Reference Harrison, Kaufman, Kaufman, Bruinicks, Rynders, Ilmer, Sparrow and Cicchetti1990), and the Columbia Mental Maturity Scales (n = 1) (Burgemeister et al. Reference Burgemeister, Blum and Lorge1972). Eleven studies used a self-reported measure of maternal depressive symptoms: the Edinburgh Postnatal Depression Scale (EPDS, n = 7) (Cox et al. Reference Cox, Holden and Sagovsky1987), the Center for Epidemiologic Studies Depression Scale (CESD, n = 2) (Radloff, Reference Radloff1977), or the Beck Depression Inventory (n = 2) (Beck et al. Reference Beck, Ward, Mendelson, Mock and Erbaugh1961). Clinician-based diagnostic measures of depression based on the Diagnostic and Statistical Manual of Mental Disorders III or IV (APA, 1980, 1995) were used to assess or confirm the exposure status for seven studies. The majority of studies reported a contrast between high postnatal depressive symptoms v. few or no symptoms with the time of measurement varying from 6 weeks to 42 months postpartum. Two studies (Husain et al. Reference Husain, Cruickshank, Tomenson, Khan and Rahman2012; Otake et al. Reference Otake, Nakajima, Uno, Kato, Sasaki, Yoshioka, Ikeno and Kishi2014) compared children based on maternal prenatal depressive symptoms reported in the second and third trimesters of pregnancy.
The meta-analysis of crude estimates from the 14 studies (Fig. 2) showed statistically significant relationships between maternal depressive symptoms and child cognitive development (Cohen's d = −0.25, 95% CI −0.39 to −0.12), indicating a −0.25 s.d. difference in the mean cognitive scores for children whose mothers had high v. low scores on measures of depressive symptoms. The heterogeneity of the findings was marginally significant (Q statistic = 22.34, p = 0.05, I 2 = 41.8%), indicating possible variability in effect sizes across studies. The funnel plot was roughly symmetric with non-significant p values from both the Egger's linear regression test (p = 0.22) and Begg's rank correlation test (p = 0.32), an indication of minimal publication bias due to the possibility that the small studies are being undervalued/represented in the analysis. The synthesis of the ten studies using BSID as the outcome measure revealed a consistent association (Cohen's d = −0.21, 95% CI −0.36 to −0.06, I 2 = 34.1%). When the meta-analysis was restricted to studies measuring postnatal depressive symptoms (n = 12), the association remained similar (Cohen's d = −0.27, 95% CI −0.43 to −0.11, I 2 = 50.3%). When restricted to studies with longitudinal designs, the association was consistent (Cohen's d = −0.27, 95% CI −0.41 to −0.14) and the heterogeneity was non-significant (Q statistic = 17.08, p = 0.11, I 2 = 35.6%). The synthesis of the seven studies with diagnostic measures for maternal depression or a combination of diagnostic and screening measures yielded consistent results (Cohen's d = −0.28, 95% CI −0.50 to −0.06, I 2 = 55.7%). To establish the temporality of the association between maternal depressive symptoms and child cognitive development, a subgroup analysis on studies which assessed maternal depressive symptoms at least 4 months earlier than child cognitive development revealed similar findings (Cohen's d = −0.31, 95% CI −0.46 to −0.17) and the heterogeneity was non-significant (Q statistic = 8.71, p = 0.27, I 2 = 19.6%). When we restricted the analysis to the studies where maternal depressive symptoms were measured during 6–8 weeks postpartum, the association was strengthened (Cohen's d = −0.40, 95% CI −0.58 to −0.22, p < 0.0005) and the heterogeneity was greatly reduced (Q statistic = 0.24, p = 0.97, I 2 = 0.0%).
Univariate associations between maternal depressive symptoms and child cognitive development.
The meta-analysis of the four studies reporting multivariate-adjusted linear regression coefficients (Fig. 3) revealed a negative association between maternal depressive symptoms and child cognitive development (B̂ = −2.78, 95% CI −5.82 to 0.27, p = 0.074]. This result indicates a difference of −2.8 MDI points in the mean cognitive score for children of mothers with high v. low depressive symptoms, controlling for additional factors such as gender, age, and income, although this finding was not significant. The heterogeneity of the findings was significant (Q statistic = 11.56, p = 0.009, I 2 = 74.0%) across the studies. When the only cross-sectional study was removed, the association was attenuated (B̂ = −1.81, 95% CI −4.64 to 1.01) and the heterogeneity of the results remained significant (Q statistic = 6.82, p = 0.033, I 2 = 70.7%). Nevertheless, if we restricted the analysis to the three studies where maternal depressive symptoms were measured during 6–8 week postpartum and excluded the study that measured maternal depressive symptoms at 12 months, the association became statistically significant (B̂ = −4.17, 95% CI −8.01 to −0.32, p = 0.03) with decreasing heterogeneity (Q statistic = 5.26, p = 0.072, I 2 = 62%).
Associations between maternal depressive symptoms and child cognitive development αreported in studies controlling for confounders.
Discussion
Our analysis of 14 studies that focused on early childhood development revealed a statistically significant univariate relationship between maternal depressive symptoms and child cognitive development, indicating that the mean cognitive score for children whose mothers had high depressive symptoms was 0.25 s.d. lower than that for children whose mothers had few or no depressive symptoms. However, since these results do not account for confounding characteristics, care should be taken in interpreting and comparing our results.
The synthesis of the four studies reporting multivariate linear regression coefficients indicated that the mean cognitive score for children whose mothers had high maternal depressive symptoms were on average 2.8 score points lower in MDI than that for children whose mothers had few or no depressive symptoms, controlling for child gender and some socioeconomic characteristics such as family income, maternal occupation or parental education. Although this finding is not statistically significant, it is in the same direction and of a similar–though slightly attenuated–magnitude as was observed in the unadjusted analysis. It is worth noting that a high degree of variability was detected among these four studies, possibly due to their individual choice of confounders, diverse measures of depressive symptoms, and different timing of the exposure to maternal depressive symptoms (ranging from 6 weeks postpartum to 12 months), which may contribute to the non-significant findings from the adjusted analysis. In addition, when the adjusted analysis was restricted to three studies that included infants 6–8 weeks of age, the result became statistically significant and the magnitude of the association increased to an average score of 4.2 points lower on the MDI for infants whose mothers reported high depressive symptoms. This result is consistent with our previous findings in the unadjusted analysis and reveals the potential for a ‘sensitive period’ for maternal depressive symptoms’ effect on cognitive development, where early post-partum depressive symptoms might impact a mother's ability to form positive attachment relationships with the child that promote later development. (Teti et al. Reference Teti, Gelfand, Messinger and Isabella1995; Bagner et al. Reference Bagner, Pettit, Lewinsohn and Seeley2010).
These quantitative findings are consistent with prior qualitative reviews that did not combine results using statistical techniques: Field et al. (1992) reported that maternal depressive symptoms affected growth and BSID developmental scores by the end of the child's first year; and Sohr-Preston & Scaramella (2006) revealed that exposure to prenatal, postnatal and chronic depressive symptoms increased children's risk for later cognitive and language difficulties. Wachs et al. (2009) reported that children of mothers with high depressive symptoms are at risk for slow cognitive development both in high as well as low- and middle-income countries. In addition, our study included recently published studies and expanded the age range of children to infants, which is complementary to the quantitative findings in the meta-analysis published in 1998 (Beck, Reference Beck1998).
There are a number of potential mechanisms through which maternal depressive symptoms can lead to compromised cognitive development in infants and toddlers (Goodman & Tully, Reference Goodman and Tully2008). Depressive symptoms such as social withdrawal and lack of sensitivity may interfere with mothers’ responsivity to their infants’ developmental needs, (Black et al. Reference Black, Baqui, Zaman, Mcnary, Le, Arifeen, Hamadani, Parveen, Yunus and Black2007; Cooper et al. Reference Cooper, Tomlinson, Swartz, Landman, Molteno, Stein, Mcpherson and Murray2009) such as missing cues for interaction and hindering their capacity to provide early learning opportunities. The negative impact of depression on maternal energy and cognitive functioning may challenge their ability to provide appropriate care (Koutra et al. Reference Koutra, Chatzi, Bagkeris, Vassilaki, Bitsios and Kogevinas2013), such as parental prevention practices (McLennan & Kotelchuck, Reference Mclennan and Kotelchuck2000) and attending to medical needs (Minkovitz et al. Reference Minkovitz, Strobino, Scharfstein, Hou, Miller, Mistry and Swartz2005). In addition to sadness, the loss of interest in daily activities may also reduce the motivation for mothers to provide opportunities for children to play or interact with the environment in ways that support cognitive growth (O'Brien Caughy et al. Reference O'BRIEN Caughy, Huang and Lima2009). Both social support and social networks have been found to be inversely and independently related to symptoms of depression in postpartum women. (Surkan et al. Reference Surkan, Peterson, Hughes and Gottlieb2006) However, withdrawal and social isolation often present with depression, which may lead to reduced social support and assistance with child care.
Maternal depressive symptoms are often compounded by co-occurring factors such as intimate partner violence (Agrawal et al. Reference Agrawal, Ickovics, Lewis, Magriples and Kershaw2014; Stewart et al. Reference Stewart, Umar, Tomenson and Creed2014), other traumatic experiences (Leserman, Reference Leserman2008; Kukihara et al. Reference Kukihara, Yamawaki, Uchiyama, Arai and Horikawa2014; Nakai et al. Reference Nakai, Inoue, Toda, Toyomaki, Nakato, Nakagawa, Kitaichi, Kameyama, Hayashishita, Wakatsuki, Oba, Tanabe and Kusumi2014) co-morbid illnesses (Kagee, Reference Kagee2008; Di Benedetto et al. Reference Di Benedetto, Lindner, Aucote, Churcher, Mckenzie, Croning and Jenkins2014), poverty (Black et al. Reference Black, Baqui, Zaman, Mcnary, Le, Arifeen, Hamadani, Parveen, Yunus and Black2007; Grantham-McGregor et al. Reference Grantham-Mcgregor, Cheung, Cueto, Glewwe, Richter and Strupp2007), food insecurity, and child malnutrition (Baker-Henningham et al. Reference Baker-Henningham, Powell, Walker and Grantham-McGregor2003). Although these factors can be considered confounders, maternal depression may be an important mediating variable between these factors and child cognitive development. Interventions that promote child cognitive development have demonstrated improvements in maternal mental health, suggesting potential for bi-directionality in these relationships over time (Baker-Henningham et al. Reference Baker-Henningham, Powell, Walker and Grantham-Mcgregor2005; Boivin et al. Reference Boivin, Bangirana, Nakasujja, Page, Shohet, Givon, Bass, Opoka and Klein2013; Singla et al. Reference Singla, Kumbakumba and Aboud2015). Finally, access to effective mental healthcare is a critical element in reducing the suffering of perinatal women and enhancing positive interaction with their children (Poobalan et al. Reference Poobalan, Aucott, Ross, Smith, Helms and Williams2007). Integration of screening for maternal depression within the context of pediatric care may also aid in the identification of women and children at risk (Dubowitz et al. Reference Dubowitz, Feigelman, Lane, Prescott, Blackman, Grube, Meyer and Tracy2007; Heneghan et al. Reference Heneghan, Chaudron, Storfer-Isser, Park, Kelleher, Stein, Hoagwood, O'Connor and Horwitz2007). In addition to screening and treatment of maternal depression as well as increasing responsive caregiving to infants, concurrent strategies to address nutritional status, access to healthcare, parental education and economic security should be considered, particularly among impoverished populations. Additional research should also be pursued to provide evidence for the causality of these relationships, for example through assessing the effectiveness of interventions on maternal mental health that may result in an improvement on children's cognitive development.
This meta-analysis was limited by the modest number of studies fulfilling eligibility criteria: 14 studies were included in the unadjusted analysis, whereas only four of them with significant heterogeneity provided enough data for adjusted analysis. The studies were published from 1986 to 2013 and varied in quality. In addition, our unadjusted univariate analysis may reflect inaccurate relationships between maternal depression and child cognitive development due to confounding. The four studies included in adjusted analysis shared some confounders such as child gender and parental education, but they differed in others and were unlikely to be fully comprehensive in capturing all potential confounders. An additional challenge in interpreting the adjusted meta-analysis is to understand the extent to which confounding variables may be explanatory or mediating factors on the pathway from maternal depression to child cognitive development. In addition, it is possible that women's depressive symptoms may be a response to children's cognitive delays (Nicholson et al. Reference Nicholson, Deboeck, Farris, Boker and Borkowski2011; Bagner et al. Reference Bagner, Pettit, Lewinsohn, Seeley and Jaccard2013), although some longitudinal studies included in this analysis found that maternal depressive symptoms preceded infant cognitive development outcomes (Hay & Kumar, Reference Hay and Kumar1995; Cornish et al. Reference Cornish, Mcmahon, Ungerer, Barnett, Kowalenko and Tennant2005; Koutra et al. Reference Koutra, Chatzi, Bagkeris, Vassilaki, Bitsios and Kogevinas2013). The transactional effects of maternal depression and child cognitive development can result in an iterative process over time, whereby children's poor cognition may have an effect on maternal depression, which in turn can further impact the child negatively. As in any systematic review, publication bias may have affected our findings; significant findings may have been disproportionately reported in the literature, though the funnel plot in our univariate analysis indicated a minimal publication bias. Moreover, we were unable to evaluate the effect of severity of depressive symptoms on cognitive development due to limited data. In addition, our conclusions were largely based on depressive symptoms using screening questionnaires. Additional research using context-appropriate clinical diagnostic measures of major depression needs to be conducted to further explore the association between child cognitive development and maternal depression. Finally, the majority of studies included in our analysis came from high-income countries, which limits the generalizability of our results to low- and middle-income countries.
In conclusion, the quantitative synthesis of research results indicates that maternal depressive symptoms are related to lower cognitive scores in early childhood. An integrated approach to advancing child cognitive development can focus on addressing maternal depression, as well as strengthening the economic situation of families, promoting responsive caregiving, and supporting early child development and nutrition programs.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S003329171600283X.
Acknowledgements
The authors acknowledge Grand Challenges Canada for funding this study as a product of the Saving Brains Project. We also thank the researchers at Harvard Medical School, Harvard School of Public Health, Harvard Graduate School of Education, Muhimbili University of Health and Allied Sciences in Tanzania, Johns Hopkins Bloomberg School of Public Health, University of Maryland School of Medicine, and Bordeaux University in France, who contributed to the analysis, writing, and formatting of the manuscript.
This study was supported by Grand Challenges Canada for a proposal entitled: Poverty-related Risk Factors for Child Development and Human Capital: Comparative National and Global Assessment (grant no. 0073-03).
Declaration of Interest
None.
