Early Child Development Outcomes of a Randomized Trial Providing 1 Egg per Day to Children Age 6 to 15 Months in Malawi

Background: Eggs are a rich source of nutrients important for brain development, including choline, riboflavin, vitamins B-6 and B-12, folate, zinc, protein, and DHA. Objective: Our objective was to evaluate the effect of the consumption of 1 egg per day over a 6-mo period on child development. Methods: In the Mazira Project randomized controlled trial, 660 children aged 6-9 mo were randomly allocated into an intervention or control group. Eggs were provided to intervention households during twice-weekly home visits for 6 mo. Control households were visited at the same frequency. At enrollment, blinded assessors administered the Malawi Developmental Assessment Tool (MDAT), and 2 eye-tracking tasks using a Tobii-Pro X2-60 eye tracker: a visual paired comparison memory task and an Infant Orienting with Attention task. At endline, 6-mo later, blinded assessors administered the MDAT and eye-tracking tasks plus an additional elicited imitation memory task. Results: At endline, intervention and control groups did not significantly differ in any developmental score, with the exception that a smaller percentage of children were delayed in fine motor development in the intervention group (10.6%) compared with the control group (16.5%; prevalence ratio: 0.59, 95% CI: 0.38-0.91). Among 10 prespecified effect modifiers for the 8 primary developmental outcomes, we found 7 significant interactions demonstrating a consistent pattern that children who were less vulnerable, for example, those with higher household wealth and maternal education, showed positive effects of the intervention. Given multiple hypothesis testing, some findings may have been due to chance. Conclusion: The provision of 1 egg per day had no overall effect on child development in this population of children, however, some benefits may be seen among children in less vulnerable circumstances. This trial was registered at clinicaltrials.gov as NCT03385252. This study was an individually randomized controlled trial (clinicaltrials.gov: {“type”:”clinical-trial”,”attrs”:{“text”:”NCT03385252″,”term_id”:”NCT03385252″}}NCT03385252) conducted in a rural area of the Mangochi district, Malawi. Eligible children were those aged 6–9 mo residing in the catchment areas of the Lungwena Health Center and St. Martins Hospital of Malindi. Details of recruitment and inclusion criteria have been published previously (5). All protocols were reviewed and approved by the Institutional Review Boards at the University of California, Davis, and the College of Medicine in Malawi. At enrollment, which occurred from February to July 2018, participants were randomly assigned to intervention or control groups in a 1:1 allocation ratio in blocks of 10. The random sequence was generated by a researcher independent of the field team. Allocation codes were concealed in sealed, opaque envelopes. After consent and a series of baseline assessments, a study staff member invited caregivers to select and open 1 envelope to reveal the child’s allocation code. The intervention consisted of 1 egg per day for the study child for 6 mo. Participants, household members, and staff members who conducted household visits were not blind to intervention group allocation due to the nature of the intervention. However, staff members who conducted developmental assessments were blind to intervention group allocation. The eggs were delivered 2 times per week by a study staff member during home visits. Each participating household was provided with a storage basket for the eggs, information about hygiene and handwashing during food preparation, recipes and suggestions for how to prepare eggs, and instructions not to share the child’s eggs with other family members. Formative research in the study communities revealed that intrahousehold sharing was highly likely, particularly with siblings. Therefore, the family was provided with an additional batch of 7 eggs per week that could be shared with other family members. During the second household visit each week, the staff person administered a 7-d morbidity history and a brief FFQ focused on animal source foods. During the first 2 wk of the trial, a study staff member visited intervention households 4 d each week to provide additional messaging, coaching, and support to promote feeding the child eggs throughout the study. These visits were repeated for 2 d after intervention households had completed 3 mo of intervention to reinforce egg preparation and safe handling messages and encourage continued adherence. The control group households were also visited twice per week and received messages about hygiene and handwashing during food preparation, but they did not receive eggs or any other foods during the study period. During the course of the trial, control households received participation incentives such as buckets and laundry tubs. At the end of the trial, they received a mixed basket of food items, including eggs. The total value of the incentives and food package matched the value of the eggs provided to the intervention households. During the twice-weekly home visits, the staff person asked the caregiver to recall the child’s most recent meal. Similar to the intervention group, morbidity histories and FFQs were also administered on the second visit each week. At a baseline clinic visit, a staff member conducted a survey asking about each child’s characteristics, administered a multipass quantitative 24-h dietary recall, and performed anthropometric measurements. Dates of birth were recorded from the child’s clinic card (95.6% of children) or parental recall. Staff members measured hemoglobin (Hb) concentrations (Hemocue 201) and tested for malaria using a rapid diagnostic test (SD Bioline Malaria Ag P.f/Pan). Children with severe anemia (Hb <5 g/dL) or positive malaria test were referred for treatment. A home visit was conducted during the following week to collect household socioeconomic and demographic information, Household Food Insecurity Access (HFIA) status (11) and administer the Home Observation for Measurement of the Environment (HOME) inventory (12). Six months after enrollment (endline), children were invited back to the study clinic and the 24-h recall, anthropometric measurements, and blood sample collection were repeated. The Family Care Indicators (FCI) interview was also administered at endline (13). All developmental assessments were conducted at the study clinic site. At baseline and endline, staff members administered the Malawi Developmental Assessment Tool (MDAT) (14) and 2 eye-tracking tasks. Participants enrolled before 4 April, 2018 were tested on the pilot version of the eye-tracking tasks at baseline (39%). After reviewing the pilot data, we revised the eye-tracking tasks and administered the final version of the tasks at the remaining baseline visits and all endline visits. The final version is described below and the pilot version in Supplemental Methods. Only data from the final version of the task were used in analyses reported here. At endline, staff members also administered an elicited imitation task. Eggs contain several nutrients that are important for global brain development, such as protein and iron (2). Therefore, we selected 1 test of global development, which was previously shown to be appropriate and valid in the local context: the MDAT (see below). Choline, also highly concentrated in eggs, plays a specific role in the development and function of the hippocampus, underlying declarative memory (15), and in the development of attentional processes (16). Therefore, we selected 2 tests of declarative memory, 1 behavioral task, and 1 eye-tracking task. We also selected an additional eye-tracking task, the Infant Orienting with Attention (IOWA) task, which reflects maturation of the neural circuitry underlying visual attention (17). The MDAT was designed to assess child development in Malawi, drawing items from several standard tests, such as the Denver Developmental Screening Tool and Griffiths Mental Development Scales, plus additional items drawn from culturally appropriate behaviors. Three subscales assess fine and gross motor and language development by direct observation of the child, and 1 subscale assesses personal-social development via the caregiver interview. The MDAT was previously validated in Malawi, with >94% of items showing high reliability (κ >0.4 for interobserver immediate, delayed, and intraobserver reliability) (14). Neurodevelopmental impairment was defined as whether the child failed 2 items or more in any 1 domain at the chronological age at which 90% of the normal reference population would be expected to pass. Using this definition, the MDAT demonstrated high sensitivity (97%) and specificity (82%) to detect children with neurodevelopmental impairment in Malawi (14). We applied this definition to determine the risk of a neurodevelopmental disorder and calculated continuous raw scores and z-scores based on published norms. The elicited imitation task measures children’s declarative memory (7). Our adaptation of the task comprised 8 items based on previously published versions of this task (18–21). Each item consisted of a set of toys and a sequence of 2 target actions. For example, for 1 item the toys were a ball and a dump-truck and the target actions were 1) put the ball in the bed of the truck and 2) dump it out. For each item, children first played with the set of toys for 30 s while the tester recorded any target actions spontaneously performed. The tester then demonstrated the sequence of target actions twice. Either immediately (4 items) or after a delay of mean ± SD of 9 ± 2 min (4 items), the child was given 2 30 s opportunities to imitate the sequence of target actions demonstrated by the tester. The items were adapted to the local context in an iterative series of 3 pilot tests. For further details, see Supplemental Methods. Children were scored on how many target actions they performed spontaneously and from memory (maximum 16 points each) and how many action sequences they performed in order (maximum 8 points). The scores show little variance at age 6–9 mo because children perform very few target actions, therefore we administered this test at endline only. The child’s mood, activity level, and interaction with the assessor during each developmental assessment was rated as positive or not positive. For details, see Supplemental Methods. All assessors were required to pass knowledge and practice-based evaluations before administering the tests and interviews. Interscorer agreement, which is the agreement between 2 data collectors independently scoring the same test session or interview, was high for the MDAT (95%), HOME inventory (89%), and elicited imitation task (90%). For further details, see Supplemental Methods. We conducted additional training for items that showed low agreement. Two eye-tracking tasks were administered: a visual paired comparison (VPC) task, based on Rose (22), and the IOWA task, based on Ross-Sheehy et al. (17). Each child was assessed using 1 of 2 eye-tracking systems, each of which comprised a laptop, an external monitor mounted on an adjustable arm, a webcam attached to the top center of the monitor, and a Tobii Pro X2–60 eye tracker with external processing unit. For further details, see Supplemental Methods. Each eye-tracking system was placed in a room in the study clinic site fitted with 4 black curtains. The curtains were open when participants entered the room and closed before starting the eye-tracking task, creating a curtained booth that blocked out visual distractions. When the curtains were closed, only the monitor was visible to the mother and child, who were seated in a chair facing the monitor (Supplemental Figure 1). The child was placed in an infant carrier worn by the mother to minimize the child moving around on the mother’s lap and moving out of range of the eye tracker. The monitor was positioned so that it was ∼60 cm from the child’s eyes. Staff members requested that the mother look away from the monitor or close her eyes, to avoid unintentionally directing the child’s gaze. A staff member monitored the mother and child during the session on the laptop screen via the webcam and reminded the mother to turn away if she started watching the screen. A dynamic image appeared before every trial to draw the child’s gaze to the center of the screen. In the NP task, this was a black cross that alternated with images of colorful toys, which were presented with a variety of sounds designed to draw the child’s attention. In the IOWA task, the central image was a bright yellow dynamic smiley face that loomed from small (0° 52’ width × 0° 57’ height) to large (4° 35’ width × 5° 9’ height) at a rate of ∼1.5 Hz and was accompanied by classical music. A staff member monitored the infant’s gaze and pressed a key to advance to the next trial when the infant’s gaze was located on the central image. The VPC task consisted of 4 trials. In each trial, the stimuli were a pair of African faces from the database reported in Strohminger et al. (23). In each trial a 20 s familiarization period was followed by a 20 s test for visual recognition memory (Figure 1A). In each period, 2 faces were presented on the left and right sides of the screen. During the familiarization period, the same face was presented on both sides. During the recognition memory period, the face presented in the familiarization period appeared on 1 side and a novel face appeared on the other side; the stimuli were reversed after the first 10 s. As is typical in eye-tracking tasks to help infants maintain attention and interest in the task in general, unrelated classical music played while the faces were on the screen (24). Order of stimuli presentation in the VPC task (panel A) and in each condition of the IOWA task (panel B). All images were presented on a gray background (RGB: 136, 136, 136). In the IOWA task, spatial cues and targets appeared 11° 45’ to the left or right of the central image. The visual attentional cue was a small black circle (0° 56’ diameter). Images were 4° 27’ (w) by 4° 7’ (h) of visual angle. IOWA, Infant Orienting with Attention; ms, milliseconds; RGB, red, green, blue; VPC, visual paired comparison. In the IOWA task, each trial consisted of the central image, a 100 ms spatial cue, 100 ms blank screen, followed by a target presented for 1000 ms (Figure 1B). Unlike Ross-Sheehy et al. (17), in which cues were presented with a 50-Hz pure tone, in our version of the task no sound was present with the cue. The targets were 96 colorful images of everyday objects, some of which would be familiar to infants and some that would be unfamiliar (e.g. strawberries, stapler, and banana). Our version of the IOWA task consisted of 3 experimental conditions and 1 control condition (Figure 1B). In the valid cue condition, the cue and the target were presented on the same side. In the invalid cue condition, the cue and target were presented on opposite sides of the central image (smiley face). In the double cue condition, 2 cues were presented, 1 on the left and 1 on the right. The subsequent target then appeared in the spatial location of 1 of the 2 cues. In the control condition, no cue was presented; the target appeared without a cue preceding it. Each child saw ≤24 trials in each of these 4 conditions, for a total of 96 trials. Half of the images in each condition were presented on the right side of the screen. In the testing sequence, VPC trials were intermixed with IOWA trials, in a fixed randomized order. Examples of VPC and IOWA trials are presented in Figure 1. The first VPC trial was followed by 8 IOWA trials (2 in each of the 4 conditions, as defined in the previous paragraph). Then, the second VPC trial was presented followed by 8 IOWA trials, the third VPC trial, 8 IOWA trials, and the fourth VPC trial. Finally, the remaining 72 IOWA trials were presented. The Tobii X2–60 eye tracker recorded the x and y coordinates of the focal point of the infant’s gaze at a rate of 60 Hz (60 times per second). A fixation is a period of time during which the eyes are focused on 1 location. This can be distinguished from eye movement, when the eyes are moving from 1 location to another. We used the Tobii I-VT fixation filter, which is an algorithm built into the Tobii software that classifies the raw eye-tracking data as fixations on various locations. We examined infants’ looking by creating areas of interest (AOIs), or regions in the display that contained relevant information. A look to an AOI was defined only when that look was classified as a fixation, not when the eyes were moving across the area. In the VPC task, we created AOIs for each face that was presented. Thus, 1 AOI covered the right side of the screen from the right edge of the central image to the right edge of the screen and from top to bottom, and a second AOI covered the mirror image on the left side of the screen. We created conservatively large AOIs to account for poor calibration accuracy, and to mimic the typical VPC data in which human observers simply record whether or not the infant looked to the left or right. We defined an NP score as the child’s total time looking at the AOI corresponding to the novel image divided by the total time looking at both AOIs combined during the recognition memory period. To ensure our conclusions were based on trials in which the infants were actually on-task, we excluded trials with <1 s of looking time during either the familiarization or recognition memory periods (11% of trials). We also calculated the child's mean fixation across familiarization periods, excluding fixations <100 ms, which are likely to be artifacts (<1% of trials). To analyze infant responses to the IOWA task, we created 3 AOIs; 1 around the central image (300 pixels by 300 pixels), 1 covering the left side of the screen, and 1 covering the right side of the screen (similar to the VPC task, described above). A response was scored as correct if the child's first fixation after the onset of the target was to the target side AOI. Trials in which infants’ first fixation after the target onset was to the opposite AOI were scored as incorrect. We determined the response time (RT) on each trial by calculating the time from the onset of the image to the first fixation in the target side AOI. We excluded any trial in which the child was looking at the central image for <200 ms before the onset of the target image (10% of trials). Because our goal was to evaluate infants’ eye movements from the central image (smiley face) to the target, it was important we only included trials in which infants were actually fixating on the central stimulus. In addition, following Ross-Sheehy et al. (17), we excluded trials with RTs <100 ms and >1000 ms (<1% of trials). RTs 100 ms are too fast to reflect infants’ response to the target, and instead reflect eye movements that began before the target appeared. RTs >1000 ms are too long to reflect infants’ response to the target, and likely represent trials in which the child was off-task. Finally, for the analysis of each condition, we included only those children with data for ≥2 trials. For example, if a child had a single RT in the invalid condition and 2 or more RTs in all other conditions, we excluded that child’s scores in the invalid condition only (5% of trials). We calculated task error by averaging percent error in the double and invalid cue conditions, based on Ross-Sheehy et al. (17). We calculated the following scores on correct trials, also based on Ross-Sheehy et al. (17). Cue facilitation reflects the degree of facilitation due to the valid cue compared with the no cue condition (i.e. how much faster do infants fixate the target in the valid cue conditions than in the no cue condition). It is calculated as the difference between the mean RT in the no cue and valid conditions, divided by the mean of the no cue condition. Cue interference is the degree of interference due to the invalid cue compared with the no cue condition (i.e. how much slower do infants fixate the target in the valid cue conditions than in the no cue condition), and is calculated as the difference between the mean RT in the invalid and no cue conditions, divided by the mean in the no cue condition. A statistical analysis plan was posted to Open Science Framework (https://osf.io/vfrg7/) before the intervention group code was broken. This plan included prespecified outcomes, covariates, and effect modifiers, as described below. All analyses were conducted using R version 3.5.0 (R Foundation for Statistical Computing). Analysis was by complete case intention to treat. Primary developmental outcomes were the MDAT norm z-scores, elicited imitation total actions recalled, VPC NP score and mean fixation during familiarization, and IOWA RT on correct trials. Secondary developmental outcomes were MDAT raw scores, prevalence of risk of neurodevelopmental disorder, elicited imitation total sequences recalled, and IOWA cue facilitation, cue interference, and task error. For continuous outcomes, we used linear regression models to estimate the mean difference between groups. For binary outcomes, we estimated prevalence ratios using regression models with a binomial distribution with a log link and prevalence differences using linear probability models. For MDAT and elicited imitation scores, we adjusted for baseline MDAT scores. For eye-tracking outcomes we adjusted for the child’s age at assessment, but not baseline eye-tracking scores due to missing baseline data for children tested on version 1. For analyses with repeated trials within participants (VPC NP, IOWA RT), we used robust SEs with participant as the independent unit. For the analysis on NP score, we included familiarization time on each trial as a covariate in all models. For the analysis on IOWA RT, we included a fixed effect of condition. For each outcome measure, we conducted a secondary adjusted analysis considering a prespecified list of additional covariates, including child sex, age, birth order, maternal age, maternal height, maternal education, maternal literacy, maternal marital status, maternal tribe, maternal occupation, religion, number of children under the age of 5 y in the household, HFIA score, housing and asset index, animal ownership, distance from home to water source, data collector, month of outcome assessment, village location, baseline child LAZ and weight-for-length z-score (WLZ), HOME score, FCI score, time of day of developmental assessment, and child’s mood, activity level, and interaction with the assessor during testing. For MDAT language score, we also considered the child’s primary language and whether the child was exposed to >1 language. For elicited imitation scores, we also considered the number of spontaneous target actions performed. For the eye-tracking scores, we also considered which of the 2 systems was used for data collection. We prescreened covariates in bivariate models to assess whether they were associated with the outcome prior to including them in the adjusted models. Covariates with a P value < 0.1 were included in the analysis. Any variables collected after baseline were only included in the models if they were not different between intervention and control groups. For the primary eye-tracking outcomes, we conducted an additional analysis controlling for the corresponding baseline scores among the subset of children tested on the final version at both baseline and endline. For further details, see Supplemental Methods. For the primary developmental outcomes, we examined the following 10 potential effect modifiers: child sex, birth order, baseline maternal age and education, baseline HFIA score, baseline housing and asset index, baseline LAZ, corresponding baseline developmental or eye-tracking score below median, HOME score below median, FCI score below median. If any interaction between the potential effect modifier and intervention group was significant at the P < 0.1 level, we further explored the pattern of the effect at various levels of the effect modifier.