Impacts of an egg complementary feeding trial on energy intake and dietary diversity in Malawi

Complementary feeding diets in low- and middle-income countries are generally inadequate to meet requirements for growth and development. Food-based interventions may prevent nutrient inadequacies provided that they do not displace other nutrient-rich foods. We conducted a randomized controlled trial in rural Malawi in which 660 children aged 6 to 9 months were provided an egg a day for 6 months or assigned to a control group. Dietary intake of complementary foods and drinks was assessed at baseline, 3-month midline and 6-month endline visits using a tablet-based multipass 24-h recall. Up to two repeat recalls were collected at each time point in a subsample of 100 children per treatment group. At midline and endline, usual energy intake from eggs was about 30 kcal/day higher in the egg group compared with controls (p < 0.0001). Compared with controls, children in the egg group were over nine times more likely to consume eggs at midline and endline. There was a comparable, but nonsignificant, greater total usual energy intake from complementary foods of 30 kcal/day at midline (p = 0.128) and 36 kcal/day at endline (p = 0.087). There also was a displacement of 7 kcal/day in legumes and nuts in children at endline (p = 0.059). At midline and endline, more than 80% of children in the egg group consumed a minimally diverse diet compared with 53% at midline and 60% at endline in the control group. This study illustrates that mothers in the egg group fed eggs to young children on a regular basis without substantial displacement of other complementary foods. We describe details of the study elsewhere (Stewart et al., 2019). Briefly, we conducted a controlled trial, randomized by individual and carried out in rural areas in Mangochi District, Malawi. Infants aged 6.0–9.9 months were eligible to participate if they lived in the catchment area, were a singleton child and had a mid‐upper arm circumference >12.5 cm and no evidence of bipedal oedema, haemoglobin concentration >5 g/dl, no acute illness or injury requiring hospital referral, no history of egg allergy and no reaction to egg during a test feeding, no history of anaphylaxis or severe allergic reaction to any substance, no congenital defects or chronic morbidity associated with growth or development or that may affect feeding and no plans to leave the area in the next 6 months. Mothers of eligible children participated in a group information session during which the study was described in detail and one‐on‐one informed consent process. Mothers who agreed to participate were asked to sign or thumbprint the consent form and informed of their right to withdraw from the study at any time. All protocols were approved by the institutional review boards at the University of California, Davis, and the College of Medicine, Malawi. Baseline data collection took place at the study clinic. Child length and weight were measured by trained anthropometrists using a length board (Harpenden Infantometer, Holtain Limited, UK) and digital flat scale for mother‐and‐baby weighing (Seca 874, Seca North America). Stunting, underweight and wasting were defined as z‐score <2 standard deviations (SDs) below the mean length‐for‐age, weight‐for‐age and weight‐for‐height, respectively, using the WHO child growth standards (WHO Multicentre Growth Reference Study Group, 2006). Malaria infection was determined by rapid diagnostic test (SD BIOLINE Malaria Ag P.f/Pan, Abbott Diagnostics, USA). Anaemia was defined as haemoglobin concentration <11 g/dl, as measured by haemoglobinometer (Hemocue 201, Hemocue Inc., Sweden). Demographic and socio‐economic data were collected by interview with the mother. Additional data on household characteristics and food security, assessed using the Household Food Insecurity Access Scale (Coates, Swindale, & Bilinsky, 2007), were collected by interview with the mother during a home visit the week following enrolment. Randomization occurred at the study clinic visit after initial data collection. Mothers selected and opened one opaque envelope to reveal the intervention group. The intervention consisted of one egg per day for the study child for 6 months. Eggs were delivered to the household twice‐weekly, and mothers were encouraged to give the egg to the study child only. To reduce the probability that the egg was shared, intervention households were provided an additional egg per day. Staff who conducted assessments were blinded to group assignment. We enrolled infants between February and July 2018, with follow‐up visits occurring during the subsequent 6 months. At 3 months, a midline visit including dietary and anthropometric assessments was conducted at participants' homes. At 6 months, participants returned to the study clinic for endline data collection. Our sample size of 331 children per group was calculated for the main trail outcome of linear growth and not for dietary differences associated with the intervention. For this secondary analysis, we are powered to detect a mean difference effect size of 0.22 SDs between groups (α = 0.05 and 1 − β = 0.80) for continuous outcomes using the 660 randomized participants. Dietary intake of complementary foods and drinks was assessed using a tablet‐based multipass 24‐h dietary recall (Caswell et al., 2019). In a subsample of 100 children per group, replicate recalls were conducted on up to two additional days for each visit time point. Mothers were asked if they were breastfeeding, though breast milk intake was not assessed. Interviewers for dietary data collection were trained over a 2‐week period on 24‐h interview technique, data recording and measurement of portion sizes. Supervisors completed shadow visits following a quality control checklist to ensure data quality over the course of the study. During the 24‐h recall, the child's mother was asked to recall all foods and drinks her child consumed between waking on the previous day and waking on the day of the recall. In the first pass, the mother was asked to report all foods consumed. In the second pass, she was asked for a detailed description of each food, the ingredients included in mixed dishes and the time her child consumed the food. In the third pass, she was asked to demonstrate the portion size of each food or drink her child consumed. Portion sizes were demonstrated using rice, modelling dough or water, according to the texture of the food or drink as served, as described by Gibson and Ferguson (2008). The mother was asked to show any portion left over, which was excluded from the estimated portion size. The interviewer measured the rice, dough or water using a 250‐ or 500‐ml graduated cylinder. In the final pass, the interviewer reviewed the information recorded in the first three passes and checked for any missed foods. The 24‐h recall interview also included questions on current breastfeeding status, use of supplements and factors that may have affected food intake. We converted dietary data to observed energy intakes as follows: We estimated the distribution of usual energy intakes from foods and drinks in total and by food group using the National Cancer Institute (NCI, 2012) approach (Tooze et al., 2010) detailed below. This method adjusts for measurement error—primarily from day‐to‐day variability in intakes—in observed, single‐day estimates of nutrient intakes using repeat recalls. To describe dietary diversity, we used food groups to assess infant and young child feeding practices as defined by WHO and partners (2008): (1) grains, roots and tubers; (2) legumes and nuts; (3) dairy products (milk, yogurt and cheese); (4) flesh foods (meat, fish, poultry and liver/organ meats); (5) eggs; (6) vitamin A‐rich fruits and vegetables; and (7) other fruits and vegetables. We also created an eighth category of snack foods and sugar‐sweetened beverages as described in Campbell et al. (2018). Any foods not falling in these eight predefined food groups were categorized as ‘other’. For the dietary diversity analyses, mixed dishes were disaggregated, and each ingredient was separately assigned to its corresponding food group. MDD was determined by consuming four or more of the seven food groups, excluding the additional category of snacks and sugar‐sweetened beverages and other, uncategorized foods. All testing was two‐sided and considered significant at the α = 0.05 level, unless otherwise specified. We used SAS version 9.4 primarily for analyses and results preparation and Stata version 14.1 primarily for data cleaning and management. Repeat recalls were excluded at baseline if they were not completed within 21 days after enrolment and prior to the first intervention and monitoring home visit; at midline if they were not completed within 10 days before or after the midline home visit; and at endline if they were not completed within 21 days before the endline clinic visit. As a result of these exclusion criteria, the number of children with repeat recalls is less than 100. A detailed statistical analysis plan was developed prior to initiating the analysis and posted publicly. We used descriptive statistics to explore child, maternal and household characteristics by intervention group and between those with and without endline data. To describe dietary patterns, we calculated the frequency of consumption and the quantity per serving (grams) for the following categories of foods: nsima (the local staple dish of stiff maize porridge); phala (soft porridge, typically prepared only for young children); fish; eggs; meat; dairy; fruit; green leafy vegetables; orange fleshy vegetables; other vegetables; potatoes, rice or pasta; sweets; savoury snacks; legumes; juice; tea; bread; and other foods. For this analysis, mixed dishes were categorized by main ingredient and, for maize‐based dishes, preparation method. To test the impact of the intervention on usual energy intake from foods and usual energy intake by food group, we used the NCI (2012) macros with bootstrap standard errors (National Cancer Institute and National Center for Health Statistics, 2014) to estimate usual intake distributions by intervention group (Tooze et al., 2010). The MIXTRAN macro fits a mixed effects model of usual energy intake, and the DISTRIB macro uses a Monte Carlo procedure to estimate mean and percentiles of the usual intake distribution. We included child age, sex, report of illness in the previous 24 h, report of unusual food intake during the recall period and whether the recall day was a market day as potential fixed effects on total energy intake or energy intake by food group. Unusual food intake and report of illness were only included in the final usual intake models if they showed a bivariate association (p < 0.1) with the outcome variable. Usual intake distributions were modelled separately by group to produce estimates of mean usual intake that appropriately allow day‐to‐day intake variation to be different in the two groups. For each variable describing usual energy intake or usual energy intake by food group, we repeated the usual nutrient intake modelling procedure 200 times, using bootstrap samples created by random draw with replacement. Usual intake was estimated for total energy, the seven WHO food categories, total energy without eggs and snack foods. Although the above were pre‐specified, we subsequently added an ‘other’ category to capture the full diet. This category consists primarily of energy from oil consumption. We used the original sample point estimates of usual intake and bootstrap standard error estimates to construct 95% confidence intervals (CIs) and calculate p‐values for difference in mean by intervention group based on unequal variance t‐tests. We tested for differences in mean dietary diversity score using negative binomial regression and for changes in prevalence of MDD using log‐binomial regression. All models controlled for baseline status and adjusted models controlled for child age, sex and report of illness or change in appetite if marginally associated with the outcome in bivariate regressions (p < 0.1). In addition, we similarly compared the percentages of children in each group who consumed foods from each of the seven WHO food groups at midline and at endline. Statistical analysis plans were determined prior to any analyses and posted at https://osf.io/vfrg7/ .