Maternal depressive symptoms are negatively associated with child growth and development: Evidence from rural India

Abstract Maternal depression has been suggested as a risk factor for both poor child growth and development in many low‐ and middle‐income countries, but the validity of many studies is hindered by small sample sizes, varying cut‐offs used in depression diagnostics, and incomplete control of confounding factors. This study examines the association between maternal depressive symptoms (MDSs) and child physical growth and cognitive development in Madhya Pradesh, India, where poverty, malnutrition, and poor mental health coexist. Data were from a baseline household survey (n = 2,934) of a randomized controlled trial assessing an early childhood development programme. Multivariate linear and logistic regression analyses were conducted, adjusting for socio‐economic factors to avoid confounding the association of mental health and child outcomes. MDS (measured using the Center for Epidemiologic Studies Short Depression Scale) was categorized as low, medium, and high in 47%, 42%, and 10% of mothers, respectively. The prevalence of child developmental delay ranged from 16% to 27% for various development domains. Compared with children of mothers with low MDS, those of high MDS mothers had lower height‐for‐age, weight‐for‐age, and weight‐for‐height z‐scores (0.22, 0.21, and 0.15, respectively), a higher rate of stunting and underweight (~1.5 times), and higher rate of developmental delay (partial adjusted odds ratio ranged from 1.3–1.8 for different development domains and fully adjusted odds ratio = 1.4 for fine motor). Our results—that MDS is significantly associated with both child undernutrition and development delay—add to the call for practical interventions to address maternal depression to simultaneously address multiple outcomes for both women and children.

Given that the weighted mean prevalence of antenatal and post-natal depression in low-and middle-income countries has been estimated at 25.3% and 19.6%, respectively (Gelaye, Rondon, Araya, & Williams, 2016), with prevalence of depressive symptoms in South Asian countries often even higher (Parsons, Young, Rochat, Kringelbach, & Stein, 2012), an understanding of these risks can contribute to understanding the barriers to improve nutrition and child development in countries where undernutrition remains a stubborn problem.
The apparently clear observational association of maternal depressive symptoms and child stature as indicated by an odds ratio (OR) for stunting of 1.4 in a meta-analysis belies a fair amount of variation across settings (Surkan et al., 2011). For example, only five of the 12 studies of stunting employed in that meta-analysis found a statistically significant relationship with maternal depression. Dissimilarities in results across countries reflect, in part, differences in inclusion of covariates and in the cut-offs used in the various diagnostic and self-reported measures of depression as well as intrinsic differences across sample populations.
Regarding child development, the relatively few studies in low-and middle-income contexts also suggest an association between maternal depression and child cognitive development, but validity of these studies is hindered by small sample sizes and incomplete control of confounding factors (Gelaye et al., 2016). Further results from disparate settings and with meaningful covariate controls are necessary to identify the conditions under which maternal depression most adversely impacts child growth and development.
This study examines the association between maternal depressive symptoms and various measures of child growth and development in a large sample drawn from an understudied population-rural India-specifically two rural districts in the central Indian state of Madhya Pradesh.
Relatively few studies that investigate this question are able to explore in a flexible manner the conditional association between maternal depressive symptoms and measures of both physical growth and cognitive development. Further, as poverty, deprivation, and poor mental health are often coincident (Das, Do, Friedman, Mckenzie, & Scott, 2007;Lund et al., 2010), a robust set of socio-economic controls are included to avoid confounding the association of mental health and child outcomes by socio-economic factors that may simultaneously determine both.
2 | DATA AND METHODS

| Data source and study population
The data were collected in the baseline household survey of a clustered randomized controlled trial. This study assesses a pilot programme implemented in Madhya Pradesh, India, which aims to improve early childhood development outcomes with the introduction The sample of intervention and control AWCs in these districts was selected based on two-stage randomization of the first treatment and control sectors followed by a random selection of three to four centres within each study sector, resulting in a well-balanced sample between treatment and control centres with respect to catchment size and prevalence of underweight children. 2 The survey first involved the complete listing of AWC catchment areas (200 enumeration areas) for eligible households (those currently with a child or children between the ages of 6 and 59 months). This population constituted the sampling frame for the household interview. Fifteen households in each AWC catchment area were randomly selected from this list. As a few catchment areas had less than 15 eligible households, 2,934 mother-child pairs participated in the survey. Mothers were informed about the purpose of the study, and oral informed consent was obtained from all participants.
Data were collected via face-to-face interviews using a structured ques-

| Child anthropometry
We examine the associations of maternal depressive symptoms (MDSs) with child growth, using both continuous z-scores based on the 2006 World Health Organization (WHO) child growth standards (de Onis et al., 2009) as well as measures of undernutrition for individuals more than two standard deviations below the medians of these standards.
Child anthropometry was obtained by trained field staff using standard methods (Cogill, 2003). As the trial assessed the impact of day care centres, the survey teams were instructed to select an index child from each household in the age range of 6-48 months. If there

Key messages
• Maternal depressive symptoms are significantly associated with child undernutrition in India, with possible mechanisms operating though poorer home environment, less engagement with children, lower use of health services, and suboptimal complementary feeding practices.
• Depressive symptoms also are associated with development delay among children <4 years, although the strength of the association depends on the domain of development considered.
• As poverty, poor mental health, and undernutrition coexist in India, it is crucial to consider maternal depression in endeavours to simultaneously address multiple outcomes for both women and children. were more than one child in this age range, the older child was considered the index child. This choice was to facilitate comparisons among children who had the opportunity to participate in day care centres-a focus of the study on integrated child development services. Children's weight was measured by Tanita weighing scales, accurate to 100 g. For height, supine length was measured for children from 0 to 23.9 months and standing height was measured for children ≥24 months old, using standardized length boards, which were precise to 1 mm. Children's weight and length/height measurements were then converted into height-for-age z-score (HAZ), weight-for-age z-score (WAZ), and weight-for-length/height z-score (WHZ), according to WHO (2010) child growth standards. Stunting was defined as HAZ below −2 z-score, underweight as WAZ < −2, and wasting as WHZ < −2 (WHO, 2010).

| Child development
Child development was measured by a parent-reported instrument, the Ages and Stages Questionnaire, Third Edition (ASQ-3), which includes five domains, namely, communication, gross motor, fine motor, problem solving, and personal social skills (Squires & Bricker, 2009). The ASQ-3 has been standardized on children aged from 1 month to 5.5 years.
Each questionnaire contains 30 items, grouped by developmental domains (each domain has six items), about a child's everyday activities.
Mothers were asked to select the most appropriate response ("yes," "sometime," or "not yet") based on the previous observations of the child's ability or inability to perform daily activities expected of the specific age group. The total age-standardized score for each domain is 60.
These scores were compared with the cut-off for the child-specific age group to identify children with developmental delay that need further assessment. The ASQ-3 has been widely used in many population and has been approved as a valid and reliable tool to screen child developmental delay (Fernald, Kariger, Engle, & Raikes, 2009;Kvestad et al., 2013). It has also been used as an outcome indicator of adverse effects of maternal depressive symptoms (Junge et al., 2017).

| Maternal symptoms of depression
MDSs were measured using the Center for Epidemiologic Studies Short Depression Scale (CES-D-10), an adaptation of the 20-question scale in common use (Radloff, 1977). The shortened set of questions has been validated in various settings (Irwin, Artin, & Oxman, 1999;Zhang et al., 2012), including India (Kumar, Nakulan, Thoppil, Parassery, & Kunnukattil, 2016). The CES-D-10 scale includes a 10-item checklist that measures symptoms of depression experienced over the past week, including feelings of bother, distraction, hopelessness, restlessness, and sleep disturbance. Each item can be scored on a scale of 0-3, yielding a range for the possible summary score from 0 to 30, with higher scores indicating increased severity of depressive symptoms. Some studies reduce 0-3 responses to yes/no dichotomies yielding a possible total score of 0-10 (Irwin et al., 1999;Kumar et al., 2016). We conduct analysis on the full score, but all reported results are robust to such a score truncation.
The CES-D-10 scale was translated into Hindi and checked for accuracy through back translation and pilot testing. Interviewers were trained on the protection of interviewee privacy and data confidentiality.

| Covariates
Household characteristics examined as control variables included household composition (household size, number of children under 5 years, number of working members aged 15-64 years old, and female headed household), socio-economic status (SES), and distance from household to the nearest AWC. The SES index was constructed using principal components analysis of variables such as ownership of a house and land, housing quality (construction materials for floor, roof, and wall), access to services (water, electricity, and sanitation services), and household assets (different types of durable goods and animals; Filmer & Pritchett, 2001;Vyas & Kumaranayake, 2006). The first factor derived from component scores was used to classify household SES into quartiles.
Maternal characteristics included age, education (categorized as no schooling, primary school-5 years of schooling, middle school-6-9 years, and high school or higher-≥10 years), and social group (tribal or other). At the child level, we adjusted for child age and sex in the analyses.
We also measure activities that may be influenced by MDS, including home environment, mother-child engagement, child feeding, and use of health services. Home environment was assessed by the presence of learning materials (number of picture books) and toys (homemade toys, manufactured toys, and household objects) and whether the child was left alone with siblings. Mother-child engagement was measured by items related to whether a mother interacted with her child in various activities (did she read books, tell stories, sing songs, play with child, name things/counted, or draw with the child) in the past 3 days. Child feeding was assessed by the number of food groups children consumed in the past 24 hours based WHO (2008) recommendations. Use of health services was measured by whether women received counselling on child health and nutrition at the AWC.

| Statistical analysis
The association of MDS with child growth and child development and also with maternal and household characteristics, home environment, mother-child engagement, child feeding, and use of health services was indicated using t tests of differences in means by MDS categories for continuous variables and chi-square tests for categorical variables.
Multivariate regression models were used to examine the relationship of MDS with child growth and child development, using these outcomes both as continuous scores as well as in terms of the presence or absence of undernutrition (<−2 SDs) or of developmental delays, adjusting for several potential confounding factors at household, maternal, and child levels as mentioned above and accounting for centres as a random effect with a cluster sandwich estimator clustered at the AWC catchment level.
Many of the previous studies collapse the CES-D-10 scores into a binary indicator with a local cut-off that represents a likely diagnosis of depression (Irwin et al., 1999;Kumar et al., 2016). However, for our purpose of understanding depressive symptoms as a risk factor, we instead examined the association between MDS and child development in a flexible manner. We first investigated the relation between MDS and child outcomes in a nonparametric manner then NGUYEN ET AL. divided the distribution of observed CES-D raw scores into deciles (Tables S1 and S2). We observed that only mothers in the highest decile of MDS had children with growth deficits that were greater than in other MDS groups. This suggested a relationship between maternal depression and physical growth of children among the mothers who report only the most severe depressive symptoms. In contrast, for child development, the lower score (capturing developmental delay) emerged with mothers at a medium level of MDS. Therefore, we decided to adopt a flexible specification with three cut-offs to capture this heterogeneity in the relation between MDS and child development: low (first five deciles, score ≤ 8), medium (next four deciles, score 9-14), and high (highest decile, score ≥ 15). We also conducted sensitivity analysis, applying the same method to collapse the CES-D-10 scores into a binary indicator with a local cut-off of ≥4 (Kumar et al., 2016), which is roughly at the lowest quartile in our sample. This cut-off includes mild, moderate, and severe depression under other criteria for diagnosis reported in that study, although cut-offs for moderate and severe depression were not reported. All statistical analyses were carried out using Stata version 14 software.

| Sample characteristics
Characteristics of the study population, overall and by the three MDS categories, are presented in Table 1. The average household size was 7, with 1.6 children <5 years and 3.7 members in working age range. The average MDS score is 9.2 (range 1-24), with 47% was categorized as low MDS, 42% as medium MDS, and 10% as high MDS.
Women with high MDS were on average, were older, had lower education levels, and were more likely to be working as farmers and from a tribal community. They also lived in households with lower SES, fewer adults, and resided a further distance from the AWC (as the AWC is generally centrally located, this is taken as measure of geographic remoteness). Furthermore, mothers with high MDS reported a poorer home environment, less engagement with children, and lower use of health services, compared with low MDS mothers. Children of mothers with high MDS also had poorer dietary diversity.

| Associations between maternal depressive symptoms and child growth
Levels of undernutrition in the study sample are high, with a rate of 62.5% stunting, 56.3% underweight, and 22.2% wasting. However, given the focus on index children, the sample is not distributed evenly across age among children 0-5 years. Because, in general, stunting increases as a child ages in the first 1,000 days, stunting rates are expected to be high compared with conventional reporting such as commonly presented from Demographic and Health Surveys. Children of mothers with high MDS had a higher rate of stunting and underweight ( Figure 1a) than their counterparts, with the odds~2 times in models adjusted for child age and sex, and~1.4 to 1.5 times in fully adjusted models (Table 2). There was no significant association between maternal MDS and wasting.  Table 3). These associations were attenuated about a half after controlling for other covariates, but they remained statistical significant at −0.22 for HAZ, −0.21 for WAZ, and −0.15 for WHZ (Table 3).

| Associations between maternal depressive symptoms and child development
The prevalence of child developmental delays as indicated in the ASQ assessment was 22% for communication, 16% for gross motor, 25% for fine motor, 21% for problem solving skills, and 27% for social skills.
Mothers with medium and high levels of MDS had a significantly  Note. Model 1 adjusts only for child age and gender. Model 2 adjusts for child age, gender, tribal caste, mother as household head, number of children <5, number of working age members, mother's age, mother's education, and household socio-economic status. All models adjust for observational dependency across observations at the Anganwadi centre catchment level.
*p < .05. **p < .01. ***p < .001. NGUYEN ET AL. 5 of 9 bs_bs_banner higher proportion of having a developmental delay child compared with mothers with low levels of MDS (Figure 2), with the partial adjusted OR ranged from 1.3 to 1.8 for different development domains and fully adjusted OR = 1.3 to 1.4 for fine motor (Table 4).
Compared with children with low MDS mothers, those with mothers that had medium and high MDS had lower developmental score in communication, gross and fine motors, problem solving, and personal social skills (Table 5). These associations attenuated in the full model after adjusting for child, maternal, and household factors and only remained statistically significant for fine motor development  Note. Model 1 adjusts only for child age and gender. Model 2 adjusts for child age, gender, tribal caste, mother as household head, number of children <5, number of working age members, mother's age, mother's education, and household socio-economic status. All models adjust for observational dependency across observations at the Anganwadi centre catchment level. HAZ = height-for-age z-score; WAZ = weight-for-age z-score; WHZ = weight-for-length/ height z-score.
*p < .05. **p < .01. ***p < .001. Note. Model 1 adjusts only for child age and gender. Model 2 adjusts for child age, gender, tribal caste, mother as household head, number of children <5, number of working age members, mother's age, mother's education, and household socio-economic status. All models adjust for observational dependency across observations at the Anganwadi centre catchment level.

FIGURE 2 Association between maternal depressive symptoms and child development
(β = −1.8 for low MDS and −3.9 for high MDS). Child development results are generally robust to an analysis that discretizes the CES-D individual questions and then adopts a cut-off of 4 as used in a previous study (Kumar et al., 2016). However, this treatment of the data fails to capture the negative relation between MDS and child growth observed at the highest decile of depressive symptoms (Tables S3 and S4).

| DISCUSSION
This paper indicates a considerable public health problem of child undernutrition and delays in other dimensions of development in rural Madhya Pradesh in India. It corroborates previous findings that MDS is significantly associated with both child growth and development.
Children of mothers with high MDS were approximately 2 times more likely to be stunted and underweight and 1.3-1.8 times more likely to have developmental delay in the models adjusting for child age and sex only. Although the inclusion of socio-economic controls roughly halves the estimated association between maternal mental health and child outcomes, significant associations with physical growth and fine motor skills persist; the OR for stunting in the adjusted model is in keeping with the existing literature (Surkan et al., 2011). We also observe that physical or cognitive developmental deficits are associated with different threshold levels of depressive symptoms. For child development, the lower score for developmental delay was observed among mothers at a medium level of MDS. In contrast, poorer child growth was only observed among the mothers with the highest decile of MDS.
The mechanisms by which maternal depression may create a risk of poor child outcomes are complex. Maternal care behaviours and practices are important elements for translating resources for care (including food, health care, psychosocial stimulation, and emotional support) into child well-being (Engle, Menon, & Haddad, 1999). Maternal depression reduces a mother's ability to take adequate care of her child, which in turn can have negative effects on the child's growth and development (Rahman, Patel, Maselko, & Kirkwood, 2008;Stewart, 2007;Walker et al., 2007). The roles of care behaviours in constrained conditions such as poverty are even more important, because good maternal care can optimize the available resources to promote children's well-being. Recent review papers have examined more comprehensive mechanisms underlying associations between maternal depression and adverse child outcomes, including both biological and psychosocial potential mechanisms from fetal programming, genetic influences, and gene-environment interactions (Herba et al., 2016), medical complications of pregnancy, the delivery of low birthweight and/or preterm infants, and environmental and nutritional influences (Gelaye et al., 2016). Our study also showed significant association between high MDS with poorer home environment, less engagement with children, lower use of health services, and suboptimal  would not allow us to distinguish the relationship of between moderate or severe MDS and child growth, which was only observed at the highest decile. To overcome this challenge, we applied flexible methods including first investigating the nonparametric association between MDS and child outcomes, which motivated the modelling choice of deciles with a twopart cut-off. However, these cut-offs may differ from other studies and thus hinder the comparison across studies.

| CONCLUSIONS
In conclusion, our study results add to the accumulating body of evidence that maternal depression is a risk factor for impaired child growth and delayed child development. Our findings echo the call for practical interventions to mitigate depression directly (Rahman et al., 2013) in maternal and child health and nutrition programmes (Bhutta et al., 2008). It is possible that social safety net or social support programmes that address the modifiable determinants of maternal depression may play a complementary or synergistic role as well (Alderman & Fernald, 2017). If such programmes are feasible, they would be intrinsically joint social support, nutrition, and development endeavours. Moreover, unlike many programmes, they could simultaneously address the woman as an individual and not merely as a caregiver.

ACKNOWLEDGMENTS
We thank the dedicated efforts of the field staff and the women who participated in the study.

CONFLICTS OF INTEREST
The authors declare that they have no conflicts of interest.