Timing, intensity, and duration of household food insecurity are associated with early childhood development in Kenya

This study examines the association between 3 dimensions of food insecurity (timing, intensity, and duration) and 3 domains of child development (gross motor, communication, and personal social). Longitudinal data from 303 households (n = 309 children) visited 9 times over 2 years were collected. Children in households experiencing severe food insecurity 3 months prior (timing) had significantly lower gross motor (β −0.14; 95% CI [0.27, −0.0033]; p =.045), communication (β −0.16; 95% CI [−0.30, −0.023]; p =.023), and personal social (β −0.20; 95% CI [−0.33, −0.073]; p =.002) Z-scores, using lagged longitudinal linear models controlling for current food insecurity; these results were attenuated in full models, which included maternal education, household asset index, and child anthropometry. Children in households that experienced greater aggregate food insecurity over the past 2 years (intensity) had significantly lower gross motor (β −0.047; 95% CI [−0.077, −0.018]; p =.002), communication (β −0.042; 95% CI [−0.076, −0.0073]; p =.018), and personal social (β −0.042; 95% CI [−0.074, −0.010]; p =.010) Z-scores; these results were also attenuated in full models. Children with more time exposed to food insecurity (duration) had significantly lower gross motor (β −0.050; 95% CI [−0.087, −0.012]; p =.010), communication (β −0.042; 95% CI [−0.086, 0.0013]; p =.057), and personal social (β −0.037; 95% CI [−0.077, 0.0039]; p =.076) Z-scores; these results were no longer significant in full models. Our findings suggest that acute and chronic food insecurity and child development are related, but that many associations are attenuated with the inclusion of relevant covariates. The study was conducted on Mfangano Island in Nyanza Province, Kenya. Lying within Lake Victoria, Mfangano Island has a population of 21,000 (Mbita Division, 2009). Broadly representative of much of the Lake Victoria region, the island’s inhabitants are vulnerable to changes in the lake as fishery involvement for trade and subsistence is widespread (Fiorella et al., 2014). Mfangano Island is rural with no running water or paved roads, and limited electricity and health services. In July 2012, households were selected with stratified random sampling based on the regions of Mfangano Island. The 4 regions bordering the lakeshore (North, South, East, and West) are defined by the Kenyan government, and 2 additional regions––a nearby island (Takawiri) and a community atop a small mountain at the centre of the island (Sokolo)––were separately defined for sampling because of geographic diversity and hypothesized differences in livelihoods. The number of households selected from each region was proportional to its population. Three hundred and three households living on Mfangano Island with at least 1 child under the age of 2 years in residence were enroled. In households with more than 1 child under the age of 2 years, the youngest child was selected to participate; in the case of twins, both were enrolled (n = 309 children, inclusive of twins). Five child deaths, including 1 twin, occurred during the study (2012–2015) and these participants were removed from analyses. Therefore, final results include 299 households and 304 children. Data were collected from December 2012 to April 2015 as part of a longitudinal panel study on fishing livelihoods, fish consumption, and child nutrition. Participants were located every 3 months for 9 visits. Local enumerators conducted surveys in Dholuo, the regional language. Surveys and data collection tools were developed from validated measures and locally adapted. The Household Food Insecurity Access Scale (HFIAS) was used to measure household food insecurity (Coates, Swindale, & Bilinski, 2007). The scale is widely utilized internationally and has been validated in several LMICs (Coates, Wilde, Webb, Rogers, & Houser, 2006; Desiere, D’Haese, & Niragira, 2015; Gebreyesus, Lunde, Mariam, Woldehanna, & Lindtjørn, 2015; Knueppel, Demment, & Kaiser, 2010; Maes, Hadley, Tesfaye, Shifferaw, & Tesfaye, 2009). The HFIAS has also been correlated with indicators of poverty and food consumption in various contexts (Becquey et al., 2010; Coates et al., 2006; Knueppel et al., 2010; Maes et al., 2009). The HFIAS and similar experience‐based measures of household food security are robust indicators of food access (Leroy, Ruel, Frongillo, Harris, & Ballard, 2015). The HFIAS includes 9 questions about food access, quantity, and quality over the previous 30 days and was administered to households every 3 months (9 times in total). Questions were scored from 0 to 3 depending on whether the respondent experienced the described condition never (0 times), rarely (1–2 times), sometimes (3–10 times), or often (>10 times). Food insecurity scores were calculated by summing each household’s responses to the 9 HFIAS questions at each time point and dividing scores by 10 for ease of interpretation. Scores at each time point range from 0 (least food insecure) to 2.7 (most food insecure, maximum possible value). Three variables assessing different dimensions of household food insecurity were created using the HFIAS scores: timing, intensity, and duration. Timing values are the raw scores at each time point. The intensity of food insecurity was calculated as each household’s cumulative score across all 9 time points over the 2‐year study, ranging from 0 (least food insecure) to 24.3 (most food insecure, maximum possible value). The duration of food insecurity was quantified as the number of times (of the 9 time points) throughout the 2‐year study that a household had a food insecurity score in the top quartile (most severely food insecure). The duration variable of food insecurity ranges from 0 to 9 times. Subscales of the Ages and Stages Questionnaire: Inventory (ASQ:I) were used to assess 3 domains of child development: gross motor, communication, and personal social (Squires, Bricker, & Clifford, 2011). Child development was assessed every 6 months at follow‐up visits during every other 3‐month intervals, for 5 times. The ASQ:I is a modified version of the Ages and Stages Questionnaire, which is widely used for the developmental screening of children under 5 years of age (Filipek et al., 2000; Kerstjens et al., 2009; Rydz et al., 2006) and has been used in LMICs (Fernald, Kariger, Hidrobo, & Gertler, 2012). The questionnaires contain a series of age‐specific items assessing the achievement of developmental milestones and tracking the child’s progress. The gross motor subscale evaluates body and muscle movement, including tasks such as standing, walking, and balancing. The communication subscale assesses language development and the use of words or sounds to express feelings. The personal social subscale reflects emotional responses and social interactions. Subscale scores ranged from 0 to 126, 0 to 130, and 0 to 158 on the gross motor, communication, and personal social subscales, respectively. As the ASQ:I questions are designed to continue being asked until a child’s threshold is met, the number of questions asked and the maximum value varies with a child’s ability and age. Consequently, we calculated continuous scores for each subscale and converted them to Z‐scores based on this population at 2‐month age intervals. The ASQ:I was translated and adapted for our study by using local culturally appropriate items, examples, and tasks, but no substantial changes were made to the original questionnaires. The 5 Ages and Stages Questionnaire subscales (gross motor, communication, personal social, fine motor, and problem solving) were piloted to assess respondent bias, maternal accuracy in reporting children’s abilities, cultural appropriateness, and feasibility according to standard procedures (Fernald, Kariger, Engle, & Raikes, 2009). Piloting entailed asking mothers about their child’s abilities and then having children demonstrate their skills. Gross motor, communication, and personal social subscale responses were most accurate, valid, and feasible to administer in the field. The fine motor and problem solving subscales were not included because they were more difficult to measure in younger infants, had the greatest maternal recall bias, and were most challenging to locally adapt. Demographics, including household size (number of members) and maternal education (categorized into: none or some primary school and completed primary school or beyond) were assessed at baseline, 12 months and 24 months. Annual measures were used as these variables were not expected to change more frequently. Household socio‐economic status was assessed with an asset index created from the first principal component of a principal component analysis (including assets such as housing composition, electricity, toilet facilities, and ownership of clock, radio, camera, computer, television, phone, refrigerator, land, livestock, solar lighting, and furniture). This approach has been shown to be a good proxy for household wealth and was correlated with consumption expenditures (Faulkingham & Namazie, 2002; Filmer & Pritchett, 1999, 2001). Anthropometric measurements, including weight and length/height for each child were taken every 3 months (9 times in total) using standard techniques (Cogill, 2003; WHO, 2006). Weight was measured using Seca digital scales (Seca 803 Digital Floor Scale; Seca Ltd., Chino, CA, USA). Length (children <24 months) was measured using infantometers (Seca 417 Mobile Infantometer; Seca Ltd., Chino, CA, USA), and height (children >24 months) was measured using stadiometers (Seca 213 Mobile Stadiometer; Seca Ltd., Chino, CA, USA). Standardized Z‐scores for length/height‐for‐age (LAZ/HAZ) and weight‐for‐age (WAZ) were calculated to assess child growth according to WHO child growth standards (WHO, 2006). Children were classified as stunted (LAZ/HAZ < −2) or underweight (WAZ < −2). All statistical analyses were conducted in Stata 14. Three regression models were constructed for each child development subscale: gross motor, communication, and personal social, with the timing, intensity, and duration dimensions of food insecurity used alternately as independent variables. The first model was comprised of child‐level covariates, including child age in months and child sex. In the second model, time‐invariant household‐level variables such as household size, assets, and maternal education were added. In the third model, child stunting and underweight status were added. Regions did not influence results; hence, they were not included in the final models. The large number of data collection rounds enabled us to replace missing values with existing data. For variables that changed between time points, such as child age and nutritional status, we used a participant's average across proximate time points to fill in missing data. These updates were necessary in less than 10% of cases for all variables. The associations between food insecurity timing and child development subscales were assessed using longitudinal linear multivariate generalized estimating equation (GEE) models. Marginal GEE models were used to compute population‐average effects allowing for longitudinal changes. Marginal GEE models were chosen over random or mixed effects models because describing how means change in populations of individuals as covariates/explanatory variables change is more relevant for policies and programmes, and GEE models allowed for the use of robust standard errors and flexible correlation structures (Hubbard et al., 2010). Correlated data (numerous visits per household or child) were accounted for using an exchangeable correlation structure. Other correlation structures produced similar results, and this structure was chosen because there were multiple measurements on the same child and each child came from an independent household. A lagged food insecurity variable was created to estimate the effect of household food insecurity 3 months ago on current gross motor, communication, and personal social Z‐scores. All models include ASQ:I scores from 4 data collection rounds (6‐, 12‐, 18‐, and 24‐month follow‐up) and corresponding lagged food insecurity scores from 3 months prior to each round, while controlling for current household food insecurity scores. The association between food insecurity intensity and duration and child development subscales were assessed with linear multivariate regression models. Food insecurity intensity and duration are cumulative measures over the entire 2‐year study and were therefore examined in relation to child development Z‐scores at the final data collection round. Stunting and underweight status are defined in these analyses as the aggregate number of times a child was stunted or underweight during the past 2 years to capture nutritional status over the study period. The study was approved by the University of California, Berkeley Institutional Review Board and the Kenya Medical Research Institute Ethical Review Committee. Adult participants provided written consent for themselves and their children prior to enrolment.