Eating and feeding behaviours in children in low-income areas in Nairobi, Kenya

Child eating and caregiver feeding behaviours are critical determinants of food intake, but they are poorly characterized in undernourished children. We aimed to describe how appetite, food refusal and force-feeding vary between undernourished and healthy children aged 6–24 months in Nairobi and identify potential variables for use in a child eating behaviour scale for international use. This cross-sectional study was conducted in seven clinics in low-income areas of Nairobi. Healthy and undernourished children were quota sampled to recruit equal numbers of undernourished children (weight for age [WAZ] or weight for length [WLZ] Z scores ≤2SD) and healthy children (WAZ > 2SD). Using a structured interview schedule, questions reflecting child appetite, food refusal and caregiver feeding behaviours were rated using a 5-point scale. Food refusal and force-feeding variables were then combined to form scores and categorized into low, medium and high. In total, 407 child–caregiver pairs, aged median [interquartile range] 9.98 months [8.7 to 14.1], were recruited of whom 55% were undernourished. Undernourished children were less likely to ‘love food’ (undernourished 78%; healthy 90% p = < 0.001) and more likely to have high food refusal (18% vs. 3.3% p = <0.001), while their caregivers were more likely to use high force-feeding (28% vs. 16% p = 0.03). Undernourished children in low-income areas in Nairobi are harder to feed than healthy children, and force-feeding is used widely. A range of discriminating variables could be used to measure child eating behaviour and assess the impact of interventions. This cross‐sectional study was conducted in seven of the 80 health facilities, which offer child welfare services and outpatient treatment for undernutrition in Nairobi. All were located in or on the periphery of major slums, where undernutrition remains a major public health problem, (Abuya, Ciera, & Kimani‐Murage, 2012; Kimani‐Murage et al., 2015). Five facilities, Mbagathi District Hospital, Kayole II Sub County Hospital, Makadara, Embakasi and Mukuru kwa Njenga Health Centre, were government run and two, Ruben Medical Clinic, Soweto PhC clinic, were faith‐based. Children aged 6–24 months were quota sampled based on the severity of their nutritional status and whether they had started treatment with ready to use therapeutic foods. Undernourished children were eligible if they had weight for age (WAZ) or weight for length (WLZ) Z scores ≤ − 2SD. Any child with WAZ and or WLZ < −3SD was defined as SAM, with the remainder defined as MAM. Severely stunted children (<−3SD LAZ) and wasted (<−3SD WLZ) and moderately stunted (LAZ between −2SD and −3SD) children were classified as undernourished, as captured with a low WAZ. However, children who had low height (<−2SD LAZ) but higher weight (WAZ > −2SD) were classified as healthy. Children were excluded if they either had medical complications such as edema (n = 2), other medical conditions such as congenital heart disease (n = 1) or cleft lip and palate (n = 1). Undernourished children were recruited between February and July 2015, with an aim to include 150 children each with moderate versus severe undernutrition and 150 on treatment and 150 not on treatment. All eligible children identified in each heath facility were included until the quota for their subgroup was fulfilled. The healthy children were recruited in a second round of data collection between July and August 2016 and were eligible if they had WAZ > −2 SD, using gender specific WHO growth charts. Healthy children were to be excluded if they had medical conditions, which required specialized care, but this did not arise in practice. Questions used to assess eating and feeding behaviours were developed, drawing on questions used in the Gateshead Millennium Study (GMS), a UK cohort study (Wright et al., 2006) supplemented by relevant questions from the Child Eating Behaviour Questionnaire (Wardle, Guthrie, Sanderson, & Rapoport, 2001) as well as behaviours observed during preliminary meal observations in low‐income areas in Nairobi (Mutoro, Garcia, & Wright, 2019). Descriptions of all the items tested are shown in Tables S1–S4. Eating behaviour was assessed using nine variables, three hypothesized to relate to appetite and avidity and six to food refusal. The variables easy to feed, loves food and easily satisfied were used to assess appetite because we hypothesized that a child who is easy to feed and loves food is likely to have good appetite. The food refusal variables, spits out food, turns head away and holds food in mouth were selected because they have previously been shown to be associated with failure to thrive (Wilensky et al., 1996) and with slow weight gain in the GMS (Wright et al., 2006). Other refusal variables were included based on meal observations carried out by our group in similar settings in Nairobi (Mutoro et al., 2019). Caregiver behaviour during meals was assessed using eight behaviours, of which four represented coercion or force‐feeding. The force‐feeding behaviours were selected based on meal observations in Kenya. Laissez faire feeding is relatively common LMICs (Dettwyler, 1989; Moore et al., 2006); to assess this, caregivers were asked how often they left their child alone when they refused to eat. There were also two questions about stress and anxiety related to feeding taken from the GMS (Wright et al., 2006) and two about whether children fed themselves during meals and snacks. Self‐feeding was assessed because studies show that self‐feeding is associated with increased food acceptance (Moore et al., 2006). These were used to construct a structured interview schedule to be administered in Swahili, after forward and back translation to ensure accurate translation. All responses were coded using a 5‐point Likert scale, which ranged from 1 (all the time) to 5 (not at all). Data were collected by the researcher and five trained research assistants. Interviews lasted between 20 and 30 min and where possible were carried out in secluded areas of the health centres. The research team aimed to take all anthropometric measurements themselves using standardized equipment, but because of lack of space in the health facilities, in most cases, the anthropometric equipment available at the health facilities were used, but the research team assisted in taking measurements to standardize the techniques used (Lohman, Roache, & Martorell, 1992; WHO, 2008). Weight was measured using a digital weighing scale (SECA 385 digital weighing Scale III) to the nearest 0.1 kg where possible. Supine length was measured to the nearest 0.1 cm using a portable Rollameter (Raven Equipment Ltd., Dunmow, UK) or a UNICEF length board. Mid‐upper arm circumference (MUAC) was measured using MUAC tapes (S0145620 MUAC, Child 11.5 Red/PAC‐50) placed on the left arm at the midpoint between the elbow and shoulder recorded to nearest 0.1 cm. We planned to examine a wider range of novel behavioural and dietary variables between three subgroups and findings from preliminary meal observations suggested large differences in interest in food between healthy and undernourished children and in the proportion becoming upset during meals (Mutoro et al., 2019). We therefore aimed for a sample size sufficient (80% power, alpha 0.05) to detect a prevalence of 15% for any behaviour in one group compared with 30% in another (relative risk = 2). This required 150 subjects in each of the three subgroups. Analyses were conducted using Statistical Package for the Social Science (SPSS) IBM Corp. Released 2010 Version 19.0. Armonk, NY: IBM Corp and Epiinfo 7.1.5.2 Statcalc. Weight and length measurements were converted into Z scores using the WHO Anthro software version 3.2.2. Children were further classified as wasted, stunted, wasted and stunted if they had WLZ or LAZ ≤ −2SD or WLZ and LAZ ≤ −2SD, respectively. Spearman’s (nonparametric) correlations were used to assess the strength and direction of interrelationships between individual child and caregiver variables and with WAZ scores, as a composite summary of the degree of stunting and/or wasting. Cronbach’s alpha was used to assess internal consistency of variables. Variables, which showed reasonable consistency were then combined to create scores, a method used in previous studies (Bentley, Stallings, Fukumoto, & Elder, 1991, Gittelsohn et al., 1998, Wright et al., 2006). Where individual variables were used, the 5‐point Likert scale was recoded into three categories: (a) all or most of the time (1 & 2); (b) sometimes (3); and, (c) rarely or never (4 & 5). Food refusal and force‐feeding scores were created by first subtracting each variable in the score with six to get an inverted value where by high scores reflected high frequency of occurrence. The mean of food refusal and force‐feeding variables was then calculated. Indices were used to assess the degree of interest in food, food refusal, force‐feeding and maternal anxiety. This was based on the assumption that children and caregivers were likely to experience these behaviours at one point during meals, but only the frequency of occurrence and the number of behaviours during meals are a likely indicator of extreme behaviour (Dettwyler, 1989). The mean of behaviours was then used to create categories reflecting high, moderate and low occurrence. Logistic regression was used to test the association between eating and feeding behaviour indices and nutritional status (healthy vs. undernourished). All children were included in descriptive analysis, but when creating scores, children with missing data were excluded. Ethics approval for the study was obtained from the University of Glasgow Ethics review committee (200140057), University of Nairobi and Kenyatta National Hospital Ethics Review committee (P651/11/2014) and the National Council for Science, Technology and Innovation (NACOSTI/P/15/9164/5185). Access to health facilities was granted by the Nairobi county and subcounty health offices.