Article Text

Download PDFPDF
Childhood overweight and maternal depressive symptoms
  1. P J Surkan1,
  2. I Kawachi2,
  3. K E Peterson3
  1. 1
    Department of Environmental Health, Harvard School of Public Health, Boston, Massachusetts, USA
  2. 2
    Department of Society, Human Development and Health, Harvard School of Public Health, Boston, Massachusetts, USA
  3. 3
    Departments of Society, Human Development and Health and Nutrition, Harvard School of Public Health, Boston, Massachusetts, USA
  1. Dr P Surkan, PhD, ScD, Exposure, Epidemiology and Risk Program, Harvard School of Public Health Landmark Center, 401 Park Drive, 4th Floor West, Rm 415, Boston, MA 02215, USA; psurkan{at}hsph.harvard.edu

Abstract

Aim: Given the rising global prevalence of overweight associated with the nutrition transition, the objective of this study was to evaluate whether maternal depressive symptoms are related to overweight in infants aged 6–24 months.

Methods: Participants in this cross-sectional study included 589 mother–child dyads from low-income urban communities in Teresina, Piauí, northeast Brazil. While adjusting for sociodemographic and biological determinants of child growth, the study assessed the relationship between mothers’ depressive symptom scores, measured with the Center for Epidemiologic Studies Depression Scale and child weight-for-height using multivariable logistic regression. Child overweight was calculated with the World Health Organization growth curves using 85th and 95th percentile cutoffs of the weight-for-height z-score (WHZ).

Results: Children of mothers with high depressive symptoms had 1.7 and 2.3 higher odds of being over WHZ cutoffs for the 85th and 95th percentile, respectively. Child age between 18 and 24 months (compared with children 6–12 months old), being low birth weight, not receiving the Family Health Programme and breastfeeding between 6 and 12 months (compared with <6 months) were other factors inversely related to at least one of the overweight indicators (odds ratio (OR) range 0.3 to 0.6). Having a mother with fewer than 8 years of education was positively associated with child overweight (OR 1.4, 95% CI 1.0 to 2.1, for WHZ >85th%).

Conclusion: Results suggest that maternal depressive symptoms are related to overweight in children aged 6–24 months.

Statistics from Altmetric.com

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

The prevalence of overweight has been rising worldwide with increasing rapidity among children and among low socioeconomic subgroups in upper-middle income developing economies.14 In the Latin American region, especially in countries such as Brazil, a shift has occurred from a high prevalence of wasting and stunting to increasing levels of overweight among the urban poor.57 Poorer countries are thus confronted with a burden of overnutrition and undernutrition, often in the same communities, and in a Brazilian study, even within families.8 9

The role of maternal depressive symptoms in child growth has received attention in the literature on failure to thrive.10 Also, its potential role in underweight and short stature has been documented in south Asia11 12 and in underweight in Jamaica, Vietnam and India.13 14 In previous analyses of our data we found maternal depressive symptoms to be related to child short stature but not underweight.15 Stunting and overweight are considered chronic conditions reflecting adverse nutritional experiences over time,16 17 which may be influenced by maternal caregiving practices.18 Therefore, it is possible that the same processes that affect malnutrition at one end of the growth spectrum may also be in play at the other. If maternal depressive symptoms affect caregiving practices such as breastfeeding and feeding interactions,19 20 they may also be related to child overweight.

In Brazil, half of the country’s poor are concentrated in the northeast21 and Demographic Health Survey Data from 1996 showed a 4.5% prevalence of overweight (>2 SD) in this region.22 In spite of the emerging problem of overweight, however, the relationship between maternal depressive symptoms and child overweight has not been examined. The aim of our study was to evaluate the association between maternal depressive symptoms and child overweight in a low-income population of infants and toddlers in northeast Brazil.

METHODS

Sample selection

Households were selected from four geographical areas consisting of nine low-income communities in Teresina, Piauí. Geographical areas were selected on the basis of similar sociodemographic and neighbourhood characteristics. Two of these areas were served by the Programa de Saúde da Familia (Family Health Programme), whereas the other two were candidate areas scheduled to receive the programme in the near future (M Vieira, Fundação Municipal de Saúde de Teresina, before data collection in 2002, personal communication). Among the services provided by the Family Health Programme are education and growth monitoring provided by a team of community healthcare workers to children under the age of 2 years and referrals to nutrition supplementation programmes for children identified as at risk. In addition, a healthcare centre is located in the neighbourhood and is staffed by a team of health professionals to foster better access to primary care.23

To identify eligible families, a door-to-door census was carried out with approximately 8000 households, from which we identified 1432 mother–child pairs in which the mother was over 15 years of age and the child was between 6 and 24 months. To locate households, we used maps from the local sewage and water company, Abastecimento de Aguas e Esgotos do Piauí, supplemented with newer areas drawn in by our fieldworkers using AutoCAD (Autodesk, Inc, 2000, San Rafael, California, USA). Maps were used to help interviewers locate the 1432 eligible families and to identify the 613 selected households. Of these 613 households, approximately one quarter of the sample (approximately 150 families) were randomly selected from each of the four geographical areas. The index child to be included in the study was chosen by a coin flip in the households in which there was more than one child eligible. When the caregiver was not home in a selected household, interviewers returned to the household five times on different days and times before the household was removed from the study.

The households sampled included 613 primary caregivers of children aged between 6 and 24 months. The analytical sample was limited to the 595 respondents who were mothers (the others were 16 grandmothers, one father and one aunt). Of those, anthropometric data were measured in 589 of their children. Logistical difficulties limited the availability of shared measuring devices used for data collection at the time of the survey and six children were unavailable when the interviewers returned to the household with the anthropometric equipment.

Data collection

In 2002, 15 trained female interviewers from Teresina and nearby cities conducted the interviews and anthropometry under the supervision of a local study coordinator. Fieldwork training lasted 2 weeks and included supervised practice interviews in the field with mothers similar to those included in our study.

We used the CMS Hanging Spring Dial Scale (Charles Morgan and Sons, London, UK; 25 kg × 0.1 kg) to weigh children and a Lightweight Infantometer (Perspective Enterprise, Portage, Michigan, USA; range 13–94 cm, accuracy 3 mm) to measure recumbent length. Anthropometry was carried out by pairs of interviewers, with each interviewer taking separate measurements on each child. Children were weighed and measured with underwear but with no other clothing or shoes. In cases in which differences between the two measurements were greater than 0.3 cm or greater than 0.1 kg, the child was measured a third time.24 25 Available measurements were averaged and used in the analysis.

Because weight-for-age also reflects tallness, high weight-for-height, eg, over 2 SD, is considered an appropriate indicator of overweight.16 The low prevalence of obesity prevented us from using this cutoff, therefore we also used a more lenient definition of overweight based on z-scores of 1.036 and 1.645, corresponding to the 85th and 95th percentiles of weight-for-length for age and sex on the World Health Organization (WHO) Multicentre Growth Reference Study curves.1 26 These cutoffs have been recommended by the Centers for Disease Control and Prevention to screen and classify “at risk” and “overweight” among children under 24 months.17 Analysis was performed using the WHO SAS program.27

Information on child birth date and birth weight was copied from the “Cartão da Criança”, a card on which medical health personnel record vaccination information for the family records. Survey items included breastfeeding duration, maternal age and education, household possessions, household type (materials used for floor, walls and roof), sanitation, income, participation in the Family Health Programme and maternal depressive symptoms as well as other topics not included here.

Mothers’ score on the Center for Epidemiologic Studies of Depression Scale (CES-D) was the main independent variable.28 The CES-D is a 20-item scale (range 0–60) of depressive symptoms occurring during the previous week. Using a standard cutoff, we categorised the CES-D by defining a score of 16 or greater as reflecting depressive symptomatology. We developed a Portuguese version of the CES-D through a rigorous process of translation and back-translation from the English version. The internal consistency reliability of our Portuguese CES-D in our low-income sample showed a Cronbach’s alpha coefficient of 0.82. This is comparable with another Portuguese version of the CES-D (alpha 0.88) from southeast Brazil that was created simultaneously with our study.29 Thirteen mothers had data missing on one item and one woman was missing data on two CES-D items. Therefore, responses to the available items were averaged and scaled for these 14 mothers.

Statistical analysis

We used conventional cutoffs or used classifications based on response distributions to code demographic variables. The variables classified were: child gender (male, female); child birth weight (<2500 g, ⩾2500 g); child age (⩾6 to <12 months, ⩾12 to <18 months, ⩾18 to <25 months); duration of any breastfeeding (<6 months, ⩾6 to <12 months, ⩾12 months); maternal educational attainment (0–7, ⩾8 years) and number of children living in household (1 or ⩾2). A sanitation scale was created from five items: using a water filter; having collected garbage; having a sewage system or toilet not connected to the sewage system but with water; presence of a running water source in the house or yard and owning a refrigerator. This scale was used as a continuous variable in which having all five corresponded to a score of 5 (high sanitation). Lower scores also reflected the sum of items available to the households, with the worst score being 0 for households with none. If information on any item was not available, it was considered missing (N  =  13). To create a measure of socioeconomic status (SES) reflecting living conditions in addition to income, we included: total household income in multiples of the Brazilian minimum wage (3  =  ⩾$R360, 2  =  $R180 to <$R360, 1  =  $R90 to <$R180, 0  =  $R0 to <$R90) with an exchange rate of approximately 2.5 Reals to the US dollar; possessing electricity, a fan, a radio and/or a television in the home (2  =  having all four, 1  =  having three, 0  =  having 0–2); house wall type (2  =  brick, 1  =  finished mud house, 0  =  unfinished mud or plastic); house floor type (2  =  ceramic, cement, or a combination of cement and ceramic, 1  =  cement or both, 0  =  mud floor) and roof type (1  =  brick or concrete, 0  =  thatched, paper or plastic). The SES/living conditions scale was modelled as a continuous variable ranging from 10 as the best conditions to 0 as the worst conditions.

Bivariate analyses were conducted to examine the association between depressive symptoms and sociodemographic predictors to weight-for-height z-score (WHZ) cutoffs corresponding to the 85th and 95th percentiles of the WHO 2005 reference curves.30 χ2 Tests of association were calculated between dichotomous growth outcomes for overweight and categorical predictors. Using our scales as continuous variables, we tested for trends in the bivariate relationship between the SES/living conditions scale and the sanitation scale with overweight. All data analysis was conducted using SAS version 9.1 (SAS Institute, Inc, Cary, North Carolina, USA). We included known determinants of growth: child gender; birth weight; age; breastfeeding duration; maternal educational attainment; sanitation scale; SES/living condition scale; number of children in household; recent child illness and participation in the Family Health Programme in multivariable models. A combined variable for recent child illness was created from three separate questions about a child’s worms, diarrhoea and fever in the past 2 weeks as reported by the mother. As not all children actually had diagnoses of worms, an answer of “don’t know” was considered missing. In this case, the variables for fever and diarrhoea were averaged, rounded and scaled to range from 0 to 3.

We used proc genmod in SAS with the interviewer included as a random effect in logistic regression models because of potential differences in interviewer styles.31 To account for censoring of breastfeeding among children who were still breastfeeding at the time of the interview, all models also adjusted for an interaction term between current breastfeeding and child age.

RESULTS

Participant characteristics

As shown in table 1, 35% and 16% of participating children had WHZ over the 85th and 95th percentiles of the WHO growth curves, respectively. Among the youngest babies aged 6–12 months, almost double the proportion had z-scores above the 95th percentile of weight-for-height, compared with children between the ages of 18 and 25 months (p = 0.04). Child who were overweight using the >85th percentile WHZ cutoff followed a similar pattern (p = 0.06). Only approximately a quarter of children with low birth weight had >85th percentile of WHZ compared with approximately a third of children weighing 2500 g or more (p = 0.08). Compared with babies born weighing 2500 g or more, babies born in the low birth weight category were approximately half as likely to be in the >95th percentile of WHZ (p = 0.1). With the 85th percentile cutoff there appeared to be a trend of higher overweight in households with better sanitation (p<0.02) and with higher SES/living conditions (p = 0.07). Receiving the Family Health Programme was associated with a 10% higher prevalence of overweight in the >85th z-score percentile, compared with children not receiving the programme. There were no substantive differences between breastfeeding duration and the prevalence of a child being over the 85th or 95th WHZ percentile (table 1).

Table 1 Bivariate descriptive statistics for the relationship of maternal depressive symptoms and demographic variables with child overweight*

Forty per cent of children with mothers having symptoms suggestive of depression had z-scores over the 85th weight-for-height percentile, whereas only 30% of children with mothers with low depressive symptoms were in this overweight category (p = 0.004). Twenty-one per cent of children of mothers with high depressive symptoms fell into the >95% WHZ category, compared with 10% of children whose mothers had few or no depressive symptoms (p = 0.0005) (table 1).

Familial and social conditions related to child overweight

As displayed in table 2, adjusting for other covariates, compared with children in the lowest age group of 6–12 months, those who were aged 18–25 months had a lower odds of z-scores above the 85th and 95th percentiles of the WHZ reference curves. Children born with low birth weight also had lower odds of having WHZ over the 85th percentile, compared with those born weighing 2500 g or more. Children of mothers with less than 8 years of education had 40% higher odds of being in the >85th percentile of WHZ compared with those with 8 years or more. Adjusting for other covariates, we observed associations between overweight indicators with the number of children in the household as well as with maternal depressive symptoms. Compared with a child living in a household with at least one other child, if the child was an only child, he/she had a higher odds of a z-score above the 85th percentile of weight-for-height. Residence in an area without the Family Health Programme was associated with lower odds of child overweight than residence in areas with the programme. Using the >95th z-score percentile cutoff, compared with children who were breastfed for less than 6 months, infants breastfed for 6–12 months had lower odds of being overweight (table 2).

Table 2 Multivariable associations between maternal depression and sociodemographic variables with child overweight

Being cared for by a mother with high depressive symptom scores (CES-D ⩾16) was associated with approximately a 1.7-fold higher odds of a child being overweight when defined as >85th percentile, compared with overweight children of mothers reporting few depressive symptoms. With the >95% cutoff, the odds of overweight was over 2 in children whose mothers had high depressive symptomatology compared with children of mothers with low CES-D scores (table 2).

What this paper adds

Maternal depressive symptoms are associated with a greater risk of overweight in low-income Latin-American children.

DISCUSSION

Our analysis suggests that maternal depressive symptoms are associated with a higher likelihood of overweight in a population of low-income Brazilian children between 6 and 24 months of age. This finding of approximately two times higher odds of overweight in these children was significant, using both the 85th and 95th percentiles of weight-for-height overweight indices. Other findings of interest included lower odds of overweight for older children (12–18 months), those born with low birth weight and residing in areas without the Family Health Programme. Having a mother with low educational attainment was associated with elevated odds of overweight.

Although the literature on failure to thrive and several studies in lower-income countries have shown a relationship between maternal depressive symptoms and child growth delay,10 11 13 15 32 we are unaware of research examining the relationship between maternal depressive symptoms and child overweight. Interestingly, a study with data from Russia, South Africa, China and Brazil indicated that stunted children aged 3–9 years are 1.7 to 7.8 times more likely to be overweight.33 In our study, stunting and overweight in children did not often occur together (with only 8% and 4% co-occurrence using the 85th% and 95th% cutoffs for overweight, respectively). Nevertheless, the fact that maternal depressive symptoms are related to both these growth outcomes underscores the importance of further research to understand the proximal behavioural determinants of these chronic nutritional conditions.

Policy implications

Results need to be confirmed with a study design supporting causal inference. If these findings are replicated in such a study, they could provide the basis for informing the design of programmes that integrate maternal services into primary care.

The cross-sectional nature of our study makes it impossible to determine a causal association of maternal depressive symptomatology with child overweight. Maternal depressive symptoms have been related to inadequate parenting behaviours and disease prevention practices.34 Inadequate interactions between depressed mothers and their children might also influence feeding practices,35 such as encouraging the child to eat nutritious food or the likelihood of breastfeeding.20 36 Depression is associated with overweight in adults37 and maternal dietary disinhibition mediates overweight in daughters.38 It is plausible that depressive symptoms influencing a mother’s own eating patterns could also affect her child’s overweight.

Breastfeeding has been associated with a lower risk of later overweight39 40 but findings are inconsistent in preschool-age children.41 In exploratory analyses, we found an association between maternal depressive symptoms and breastfeeding duration. When the breastfeeding variable was excluded from the model, however, we found no substantial change in our estimates (data not shown.)

Low social support has been linked to higher depression in multi-ethnic mothers of children under 24 months of age.42 Depressive symptoms may hinder low-income mothers’ ability to get the resources and support they need in order to overcome barriers, such as transportation to obtain healthy foods at affordable prices.43 Alternatively, a depressed mother may be more inactive herself44 45 and engage in less active play with the child,46 47 thus affecting the child’s overall energy expenditure. Maternal depressive symptoms may affect feeding and non-feeding interactions.10 Our study was limited by its inability to test these possible mechanisms.

An advantage of this study is the rich information available on many covariates important to child growth and overweight. We have very good measures of SES/living conditions and sanitation. The SES/living conditions measure included parameters not focused exclusively on income but on household conditions and a wide range of possessions. Nonetheless, because food insecurity may be more relevant than SES/living conditions, the fact that we could not control for this directly may have resulted in residual confounding. The sanitation measure included varied aspects of sanitation (eg, use of a water filter, sewage, ownership of a refrigerator, etc).

Given that we were limited by a cross-sectional design, this study represents preliminary work that needs to be replicated. We lacked information about maternal body mass index. It is possible that maternal body mass index may confound the association between maternal depressive symptoms and child overweight but it is equally possible that maternal overweight is on the causal pathway between the two. Nonetheless, because of the cross-sectional nature of our study, even with information on maternal weight we would be unable to distinguish between these two scenarios. We did not adjust for immunisation, which could explain some variation in overweight by reducing infections; however, we did adjust for recent illness (worms, diarrhoea and fever). A limitation of these data is that we have only short-term rather than chronic measures of recent illness. Without laboratory testing for parasites, it is likely there was some misclassification and therefore incomplete adjustment for confounding. Assuming that maternal depressive symptoms are related to children having more worms, diarrhoea or fever that result in weight loss, residual confounding of these variables would be likely to result in an underestimate of the association between maternal depressive symptoms and child overweight. Also, although we believe only a small fraction of our study participants were likely to be receiving treatment for depression, we lacked information from medical records or elsewhere to adjust for this.

Our study showed an approximately twofold association between maternal depressive symptoms and child overweight for weight-for-height over the 85th and 95th percentile of reference growth curves. To our knowledge, this study is the first to report an association between maternal depression and overweight in toddlers. Only recently has research documented secular increases in overweight in infants and children under 2 years,48 despite other work showing that early weight gain predicts obesity in adulthood.4951 Given the global surge in child obesity and the recognised importance of psychosocial care on child growth and development,18 52 illuminating the role of maternal depression is important to understanding the current epidemic.

Acknowledgments

The authors would like to thank Dr Louise Ryan for her statistical advice leading up to and contributing to this analysis. They would also like to thank the participating families and data collection team, in particular Lina Maria Carvalho Vieira for supervising the interviewers.

REFERENCES

Footnotes

  • Funding: This project was supported by a Sheldon Fellowship from the Committee on General Scholarships at Harvard University and the Rockefeller Center for Latin American Studies, which provided travel grants and support for data collection.

  • Competing interests: None.

  • Ethics approval: This study was approved by the Human Subjects Committee at Harvard School of Public Health.