Effect of mother’s education on child’s nutritional status in the slums of Nairobi

Research shows that there is a strong linkage between maternal education and children’s health. Children born to educated women suffer less from malnutrition which manifests as underweight, wasting and stunting in children. Maternal education has been associated with nutrition outcomes among children in studies in various settings including Jamaica [15]; Bolivia [16]; and Kenya [17, 18]. However, the mechanisms that link mother’s education and child health in general are still not well understood. Glewwe (1999) highlights three links through which education may affect child health [19]. First, formal education of mothers directly transfers health knowledge to future mothers. Second, the literacy and numeracy skills that women acquire in school enhance their ability to recognise illness and seek treatment for their children. Additionally, they are better able to read medical instructions for treatment of childhood illness and apply the treatment. Third, increased number of years in school makes women more receptive to modern medicine. Other studies have found a strong link between maternal education, social economic status and child nutritional status. This is because educated women are more likely to get steadier, higher paying jobs; to get married to men with higher education and higher income; and to live in better neighbourhoods, which have influence on child health and survival [16, 20, 21]. Studies have also found an association between maternal education and maternal depression [22], while maternal depression has been associated with poor child health outcomes, including poor nutritional outcomes [23]. Against this background, this study focuses on the urban poor in Kenya, a country that still registers a relatively high child mortality rate in Africa (74/1000 in 2008–09) [24]. Progress towards the 2015 Millennium Development Goals (MDG) targets has been slower than anticipated, the reason why there is need for further research into ways in which maternal education influences child nutrition. The purpose of this study is to better understand factors that influence poor health outcomes in the urban slums of Nairobi, Kenya. We seek to answer the question: Does mother’s education affect child nutritional status in the context of urban poverty?

The study was carried out in two urban informal settlements of Korogocho and Viwandani in Nairobi, Kenya. The study was nested within the Nairobi Urban Health and Demographic Surveillance System (NUHDSS) run by the African Population and Health Research Center (APHRC). The NUHDSS follows a population of slightly more than 60,000 people; 57 % and 43 % from Viwandani and Korogocho slum respectively. The follow up comprises of a systematic quarterly recording of vital demographic events including births, deaths and migrations occurring among residents of all households in Korogocho and Viwandani since 2003. Other data that is collected includes: household assets, morbidity, and education which is also collected and updated regularly. The two slum areas are densely populated, and characterized by high unemployment rates, lack of a basic infrastructure, poor housing, violence, insecurity, and poor health indicators [8, 25]. The two slums are heterogeneous particularly with regards to socio-economic status: Viwandani has relatively higher levels of education and employment, high mobility of residents and split households while the population in Korogocho is more stable with greater co-residence of spouses.

This paper draws on data from the maternal and child health component of a broader longitudinal study entitled Urbanization, Poverty and Health Dynamics (UPHD) in sub-Saharan Africa, nested within the NUHDSS. The UPHD project addresses key health consequences of rapid urbanization and growing urban poverty at different stages of the life course namely childhood, adolescence, adulthood, and old age. Initially, all women who had given birth from September 2006 to January 2007, as well as their children were enrolled in the study starting in February 2007. Data were collected at baseline and updated every 4 months, at which point more mother and child pairs were enrolled. Anthropometric measurements for each child were taken during every update. Length for children aged less than 24 months was measured using a measuring board (Shorr Productions, LLC, Maryland, US) while height for those aged 24 months or more was measured using a height board (Seca 213, Seca GmbH, Germany). During each visit field interviewers administered questionnaires to collect data on vaccination, health care seeking and child care practices. In this study we utilize cross-sectional data of child stunting collected in the last round of data collection conducted between October 2009 and January 2010. The study involves eight sub-cohorts of children (n = 5156) aged 0–42 months enrolled between February 2007 and January 2010.

The Urbanization, Poverty and Health Dynamics study was approved by the Ethical Review Board of the Kenya Medical Research Institute (KEMRI). The field workers were trained in research ethics and obtained informed consent from all respondents. The NUHDSS has also been approved by KEMRI’s Ethical Review Board. Verbal consent is routinely obtained from all the NUHDSS respondents.

The analysis focuses on the effect of mother’s education on child stunting. The dependent variable is stunting; measured by height-for-age z-scores (HAZ). The height for age z-scores were generated using the World Health Organization (WHO) 2006 growth standards with the WHO Anthro 2005 program, Beta Version (World Health Organization, 2006). The variable was coded as “1” (stunted) if HAZ was < −2, and “0” (not stunted) if HAZ was ≥ −2. Mother’s education is the main predictor variable. The two education categories (primary education or less and secondary or higher) were coded as “1” and “0” respectively. The covariates included: child level characteristics, maternal demographic characteristics, maternal health knowledge and seeking behaviour; household characteristics and community characteristics.

The analysis was restricted to the most recent update for each child. Cross tabulation and univariable analysis was used to explore the relationship between mothers’ education and child stunting as well as the effect of other covariates. Cross tabulations were used to show the distribution of the various characteristics listed above among the stunted children. In addition, univariable analysis was done in an exploratory manner to determine the effect of each factor on child stunting. The univariable models were estimated using the binomial logistic regression.

Thereafter, we conducted a multiple logistic analyses. In the multiple logistic analyses we entered all the variables that were used in the univariable analysis. However, the explanatory variables were entered sequentially in groups. The key predictor (mother’s education) was entered into the model first; followed by child characteristics, maternal demographic characteristics, maternal health knowledge and healthcare seeking behavior, household characteristics and community level factors. We adjusted for clustering at the household level, to account for the nature of the data where some mothers had more than one child. Some explanatory variables had missing values for example child birth weight (34 %), maternal education (~ 18 %), ante-natal visits (~ 5 %), and social-economic status (~ 10 %). Imputation was done to address the missing values. We note that the household economic status was collected longitudinally in 2006, 2007, 2008 and 2009. In order to impute the missing data for social- economic status variable, we first used interpolation method to fill in the gap by computing the mean value for the existing data and substituting it for the missing household SES value. For the remaining missing data (from households that did not have SES information at any time-point (about 10 %), we used regression based imputation method to fill in the gaps.

Specifically, to generate the missing values for the variable with missing data, we used the following STATA commands to perform multiple imputation: mi impute chained (regress) for the continuous variables like child weight (34 % missing), mother’s age (about 17 % missing) and household SES level (before categorizing SES scores into three levels); mi impute logit for the categorical variables like mother’s education level (18 % missing) and the number of ante-natal visits (about 5 % missing); and mi impute mlogit for multi-category variables like ethnicity (16 % missing). The three multiple imputation methods are implemented in STATA version 12 based on different approaches used to perform multiple imputation for continuous and categorical variables [27‐30]. Note, the implementation of the three set of multiple imputation commands require prior execution of the following two commands mi set and mi register which prepares and informs STATA to initialize and perform multiple imputation.

We also tested for the effects of imputation by generating a binary (0/1) variable representing unimputed/imputed data and including it into the model. The coefficient for the variable was not significant, an indication that the imputation process did not change the results.

Table 1 presents percentages of stunted children by different explanatory variables. The results show that 44 % of boys are stunted compared to 34 % girls. Korogocho slum has slightly more stunted children with about 41 % compared to Viwandani with 38 %. Stunting is more common among children born to single mothers and older mothers, with 45 % and 46 % respectively. Mothers that have at most primary level of education have 43 % of their children stunted compared to 37 % for mothers with at least secondary level of education. About 35 % of children born to mothers who wanted to get pregnant at the time of conception were stunted compared to 44 % of children born to mothers who wanted the pregnancy later, and 42 % of children born to mothers who did not want to get pregnant at all. Overall, 71 % of children were born in a health facility while 26 % were born at home. Among those that were born in health facilities, 37 % were stunted compared to 45 % and 35 % who were born at home/TBA and en route to hospital respectively. Mothers with three and more births had 43 % of their children stunted compared to 41 % of mothers with two births, and 35 % among mothers with one birth. 62 % of children who had low birth weight (less than 2500) were stunted compared to 36 % of the children who were of optimal weight (above 2500).

The univariable results presented in Table 2 show that mothers education, child gender, birth weight, place of birth, slum site, mothers’ marital status, pregnancy intentions and social economic status can significantly determine a child’s nutritional status (p < 0.05). Additionally, being the first born, a female of normal weight, born in a health facility, by a mother who is currently in union, who has at least secondary education, in a household that has high SES, will be protective against child malnutrition (p < 0.05). Moreover, children born in Korogocho have a high tendency to be stunted compared to children born to mothers who live in Viwandani. However, mothers’ age, number of ante-natal visits and ethnicity do not determine stunting.

Table 3 presents the multiple regression results predicting the effect of mother’s education on child stunting in the urban slums of Nairobi. Six models were fitted sequentially in this estimation, with an aim of observing the impact of mother’s education on child’s stunting and whether the impact of education is sustained when other variables (mothers’ and child characteristics, household, and community level) are introduced into the model. The estimation started with a simple model—mothers’ education regressed against child stunting. Thereafter, we sequentially added other grouped variables. Model 1 is the baseline model which introduces one explanatory variable, mothers’ education, model 2 introduces the child characteristics, while model 3 adds mother’s demographic characteristics. Model 4 adds variables related to the mothers’ health knowledge and health seeking behaviours. Model 5 and 6 add household and community characteristics respectively.

Model 1 indicates that child stunting is significantly related to the mothers’ education level (p < 0.01). The odds of child stunting are 29 % higher for mothers with no education or lower than secondary education, relative to mothers that have at least secondary education. Addition of child characteristics (child birth weight and gender) in model 2, do not change the significance of the relationship between mothers’ education and child stunting. However, both child birth weight and gender are strongly related to stunting (p < 0.01, respectively). Males have 57 % higher probability of being stunted compared to girls, while children of low birth weight are close to three times more likely to be stunted relative to children of normal birth weight.

Model 3 introduces mothers’ demographic characteristics (mothers’ age, marital status, parity and ethnicity). The relationship between mother’s education and child stunting was only slightly attenuated by the presence of these demographic characteristics. However, mothers’ marital status, parity, and ethnicity are independently associated to child stunting. The odds of stunting for children born to mothers who were never married are 56 % higher relative to those who are currently in union respectively (p < 0.05). Mother’s parity was also marginally associated with stunting. The odds of stunting for children born to mothers who have two births, and three or more births are 31 % and 39 % higher compared to those that have one child respectively (p < 0.05). Model 4 controlling for both mother’s health knowledge and health-related behavior (family planning intentions, ante-natal visits, place of delivery and knowledge on when weaning should be initiated) had little effect on the relationship between maternal education and child stunting. However, the timing of pregnancy (whether child was wanted at conception or later or not wanted at all) and place of delivery independently determine the nutritional status of a child. The odds of stunting for a child born to a mother who gives birth in a TBA facility or at home are 39 % higher compared to giving birth in a health facility (P < 0.05). Those children whose mothers wanted to get them later had 38 % higher odds of being stunted compared to those children whose mothers wanted them at conception (P < 0.05). The association between knowledge on when to initiate complementary feeding and child stunting was not significant (p > 0.1).

Inclusion of the household socioeconomic status (SES) in model 5 marginally reduces the significance of the relationship between mothers’ education and child stunting. However, being in middle SES category does not have significant effects on child’s stunting but being in highest SES category reduces the chances of stunting among children. The introduction of SES reduces the education effect on child stunting from 1.24 to 1.22, which is still significant at 5 % significance level. Consequently, controlling for household socioeconomic status and other demographic characteristics, results in the odds for child stunting being 22 % higher for mothers who have no education or lower than secondary level education relative to those who have at least secondary education level. Introduction of slum of residence in model 6 leads to a reductionon the effects of mother’s education from 22 % to 21 %, but mother’s education is still significantly associated to child stunting. Slum of residence is not significantly related to child stunting. Separate models were run for community level variable(s), (Korogocho and Viwandani). The results gave almost the same effects and none of the model had significant effects of education on stunting. In addition, the interaction term between education and slum site for the combined data was not significant.

It is important to note that most covariates had high level of correlation. To account for the correlation, we computed the significance of the correlation among the variables. Results indicate that the correlation levels among covariates are high and significant (significance level below 0.05) except correlations between the sex of the child and other covariates. The observed high correlation among covariates has effects on the fitted logistic regression models. For example, at univariate levels the site (Viwandania and Korogocho) was significantly related to stunting but at multivariate level, the variable (site) was no longer significantly related to stunting. This can be explained by high correlation that exist between the site and other variables like marital education level and age, place of delivery and social economic status of a household.