Nutrition as an important mediator of the impact of background variables on outcome in middle childhood

Adequate nutrition is fundamental to the development of a child’s full potential. However, the extent to which malnutrition affects developmental and cognitive outcomes in the midst of co-occurring risk factors remains largely understudied. We sought to establish if the effects of nutritional status varied according to diverse background characteristics as well as to compare the relative strength of the effects of poor nutritional status on language skills, motor abilities, and cognitive functioning at school age. This cross-sectional study was conducted among school-age boys and girls resident in Kilifi District in Kenya. We hypothesized that the effects of area of residence, school attendance, household wealth, age and gender on child outcomes are experienced directly and indirectly through child nutritional status. The use of structural equation modeling (SEM) allowed the disaggregation of the total effect of the explanatory variables into direct effects (effects that go directly from one variable to another) and indirect effects. Each of the models tested for the four child outcomes had a good fit. However, the effects on verbal memory apart from being weaker than for the other outcomes, were not mediated through nutritional status. School attendance was the most influential predictor of nutritional status and child outcomes. The estimated models demonstrated the continued importance of child nutritional status at school-age. © 2013 Kitsao-Wekulo, Holding, Taylor, Abubakar, Kvalsvig and Connolly. The study was cross-sectional in nature. The study was conducted in Kilifi District, Kenya, among a predominantly rural community. The majority (66.8%) of the population lives below the poverty line and is therefore unable to access basic needs due to geographical, economic, and sociocultural barriers (Kahuthu et al., 2005). The district is a food deficit region relying on trade with other districts to meet the food gap—however, income-generating opportunities are few and unsustainable (FAO Kenya, 2007). Malnutrition remains rampant due to variability in crop production; and high illiteracy levels increase the population’s vulnerability to food insecurity [Kenya National Bureau of Statistics (KNBS) and ICF Macro, 2010]. Children between the ages of 8 and 11 years were recruited from the catchment areas of five local primary schools distributed across neighborhoods ranging from sparsely populated rural areas to more densely populated semi-urban areas. The total sample of 308 children comprised both schooling and non-schooling children. Their first language was Mijikenda, the local vernacular or Kiswahili, the lingua franca and national language. The Ten Questions Questionnaire (Mung’ala-Odera et al., 2004) was administered to parents to determine the presence of any impairments or serious health problems in children. When the parent was not able to determine if the child had any impairments (visual, auditory, or motor) or in cases where only milder concerns were reported, testing was attempted. Children who were physically unable to perform the tasks were excluded. The Kenya Medical Research Institute/National Ethics Review Committee (KEMRI/NERC) provided ethical clearance for the study. Permission to visit schools was obtained from the District Education Office. We explained the purpose of the study to the head teachers of selected schools and then sought their permission to recruit children. We also held meetings with community leaders, elders, and parents (and guardians) of selected pupils to explain the purpose of the study. After each meeting, a screening questionnaire was administered to establish if selected children met the study’s eligibility criteria. We presented information on the study to parents in the language with which they were most familiar. We then obtained written informed consent for their children’s participation. All the selected children assented to their participation in the study. Building on the extant research literature, our analysis included age, gender, area of residence, school attendance and household wealth as underlying biological and environmental influences, nutritional status as a mediating variable and language skills, motor abilities and two factor scores of cognitive function as child outcomes. In order to test the various hypothesized relationships, we developed the model presented in Figure ​Figure11. Hypothesized model for testing the mediating influence of nutritional status on child neurocognitive outcomes. In the full model which included all the explanatory variables, the use of structural equation modeling (SEM) allowed the disaggregation of the total effect of the explanatory variables into direct effects (effects that go directly from one variable to another) and indirect effects (effects between two variables that are mediated by at least one intervening variable) (Bollen, 1989). We hypothesized that the effects of area of residence, school attendance, household wealth, age, and gender on child outcomes are experienced directly. Additionally, we hypothesized that the influence of these variables has an indirect effect on child outcomes through their influence on nutritional status. The model also took into account possible correlations among the five background variables. We fitted separate models for language skills, motor abilities, verbal memory, and executive function to see if there were differences among the four child outcomes. Information on child gender, age, school attendance (number of years that child has attended school), and household wealth was collected using a standard questionnaire. Birth records were used, where available, to confirm the child’s date of birth. In the cases where records were not available, the procedure outlined by Kitsao-Wekulo et al. (2012) was followed. For the purpose of this study, an age variable in 6-month increments was created. An index of household wealth that divided the sample into three approximately equal groups—least wealthy (Level 1), moderately wealthy (Level 2), and the most wealthy (Level 3)—was derived from six socioeconomic indicators: maternal and paternal education, maternal, and paternal occupation, type of windows in the child’s dwelling and ownership of small livestock. Area of residence was characterized as rural or peri-urban according to the most common settlement within the school catchment area. Children’s heights were measured to the nearest centimetre using a stadiometer and height-for-age indices were calculated using EpiInfo (Centers for Disease Control, Atlanta, GA). Growth retardation was defined as height that was more than 2 standard deviations below levels predicted for age according to the World Health Organization reference curves for school-aged children (World Health Organization, 2007). A battery of neuropsychological tests was used to assess children’s language skills, motor abilities, and cognitive functioning. Language skills. The Kilifi Naming Test (KNT), a test of confrontation naming, was used to assess expressive vocabulary (Kitsao-Wekulo et al., in preparation). In the KNT, the child was asked to spontaneously give one-word responses when presented with a black and while line drawing of a familiar object. Correct responses were coded “1.” A stimulus cue was provided when no response was given, the child stated that they did not know the name of the item or the item was perceived incorrectly. If the child did not provide a correct response after the stimulus cue, the word that was provided was recorded verbatim. The test was discontinued after six incorrectly named consecutive items. The final score was calculated by summing the number of spontaneously correct items and the number of correct items following a stimulus cue. These scores were standardized enabling the direct comparison of children’s performance across tests. Motor abilities. Children’s motor abilities were assessed using five tests of gross motor abilities covering two areas of motor performance—static and dynamic balance—and three timed tests of fine motor coordination and manual dexterity (Kitsao-Wekulo et al., under review). Age-corrected scores were obtained by computing differences between observed and predicted scores in units of standard error of the estimate (i.e., in z-score units). Maximum likelihood factor analysis with oblique rotation was then applied to the z-scores to reduce the multiple motor scores to ability composites (Ackerman and Cianciolo, 2000). Factor analysis yielded support for a two-factor solution; four tests loaded on the Motor-Co-ordination factor while the remaining four tests loaded on the Static and Dynamic Balance factor. Factor scores were defined as the mean of the z-scores for the tests loading on each factor. An Overall Motor Index was defined as the mean of the two factor scores. Cognitive functioning. We administered eight tests of cognitive functioning. These included: A detailed description of the tests is presented elsewhere (Kitsao-Wekulo et al., 2012). To reduce the test battery to a smaller set of ability composites, z-scores for each measure were subjected to principal component factor analysis with Varimax rotation. Based on factor content, skill composites were labeled Executive Function and Verbal Memory. Skill composites of the z-scores comprising each factor were computed based on factor weightings. All the tests were administered at a school near the child’s home. Each child was tested individually in a quiet area within sight of other children, and in familiar surroundings to minimize test anxiety. Observations by the assessors suggested that none of the children was unduly anxious during the test sessions. Independent samples t-tests, Chi-square tests and univariate analysis were undertaken to determine group differences in nutritional status and outcomes. Pearson product-moment correlation coefficients were used to examine the relationship between the background variables and cognitive outcomes, language skills, motor abilities, and nutritional status. AMOS version 20 (SPSS) was used to test the fit of the overall model and to examine the relationships among the variables. SEM was used to examine the relationships between background characteristics, child nutritional status and child outcomes. We developed and tested a path analysis model (Figure ​(Figure1)1) based on logic and theory about how background variables co-vary with nutritional status, and how they influence child outcomes directly and indirectly. In the full model which included all the explanatory variables, this format allowed us to test the mechanisms through which each of the background variables influenced various child outcomes directly and indirectly though a mediated path. An independent disturbance term that represented unexplained variance was estimated for each endogenous variable. In fitting the Structural Equation Models, missing information was taken into account using the Maximum Likelihood (ML) Estimates. The ML technique assumes data are missing at random for continuous, binary, and categorical variables. All direct and indirect paths were tested and each of the four child outcomes was analyzed in isolation. Specific procedures for model development were to remove non-significant paths (p = 0.05) and use modification indices as suggested by the AMOS SEM program (Arbuckle, 1988) to add paths or correlations that would improve model fit. Chi-square analysis was conducted in initial examination of the goodness of fit to insure non-significance. However, because this method is sensitive to sample size, other indices of goodness of fit included the Tucker Lewis Index (TLI), Comparative Fit Index (CFI), and Root Mean Square Error of Approximation (RMSEA) (Bentler and Chou, 1987; Browne and Cudeck, 1993). Acceptable fit was defined as TLI and CFI >0.90 and RMSEA <0.08 and an excellent fit as TLI and CFI >0.95 and RMSEA <0.05.