Associations between individual variations in visual attention at 9 months and behavioral competencies at 18 months in rural Malawi

Theoretical and empirical considerations suggest that individual differences in infant visual attention correlate with variations in cognitive skills later in childhood. Here we tested this hypothesis in infants from rural Malawi (n = 198–377, depending on analysis), who were assessed with eye tracking tests of visual orienting, anticipatory looks, and attention to faces at 9 months, and more conventional tests of cognitive control (A-not-B), motor, language, and socioemotional development at 18 months. The results showed no associations between measures of infant attention at 9 months and cognitive skills at 18 months, either in analyses linking infant visual orienting with broad cognitive outcomes or analyses linking specific constructs between the two time points (i.e., switching of anticipatory looks and manual reaching responses), as correlations varied between -0.08 and 0.14. Measures of physical growth, and family socioeconomic characteristics were also not correlated with cognitive outcomes at 18 months in the current sample (correlations between -0.10 and 0.19). The results do not support the use of the current tests of infant visual attention as a predictive tool for 18-month-old infants’ cognitive skills in the Malawian setting. The results are discussed in light of the potential limitations of the employed infant tests as well as potentially unique characteristics of early cognitive development in low-resource settings. 444 infants without known congenital malformation, severe illness, or visual impairment were enrolled after birth into a prospective cohort study in Lungwena and Malindi areas, Mangochi District, Malawi [32]. Recruitment was stratified based on infants’ gestational age at birth to enroll infants born preterm (32.0–36.9 gestational weeks), early term (37.0–38.9 gestational weeks), and full term (39.0–41.9 gestational weeks). Infants took part in eye tracking tasks at the chronological age of 9 months (±14 days) and development assessments at the age of 18 months (±1 month). Anthropometrics and background data were collected between the enrollment and 18 months of age. We conducted the study in accordance with the ethical standards of the Helsinki declaration. The study protocol was approved by the College of Medicine Research and Ethics Committee, Malawi; the Ethics Committee of Pirkanmaa Hospital District, Finland; and the Ethics Committee of the Tampere Region, Finland. A written informed consent was obtained from a parent or legal guardian on behalf of the participants at the enrollment and before the 18-month-visit. Infant’s cognitive development at 9 months of age was measured with three eye-tracking-based tasks (Fig 1) generating measures in four domains of early attentional capacities: visual search latency, visual search in the context of interfering stimuli, anticipatory attention shifts, and attention to faces. Details of these assessments are provided in Pyykkö et al. [32]. a) Three conditions of the visual search task. b) Sequence of the anticipatory attention shifts task. c) Sequence of the attention to faces task. Infants were seated in front of a 22-inch monitor which displayed the tasks and their gaze was tracked with a remote Tobii X2-60 eye tracker (Tobii Technology, Stockholm, Sweden). Gaze data were recorded on 60 Hz and consisted of the onset times of images, xy-boundaries of active areas of interest on the screen, and xy-coordinates with validity estimates of the participants’ point of gaze. After calibration, the three tasks were performed twice with a break between two sessions. Calibration had five points (cartoon images in four corners and the center of the screen) which appeared one at a time, after the participant moved their gaze into one. Calibration was done a maximum of three times to achieve a satisfactory calibration. Assessor rated the calibration as “good”, “OK”, “poor”, or “invalid” by comparing the visualization of the calibration outcome to predefined criteria. In the visual search task (based on [26]), we measured infants’ reaction time to move their gaze to a salient visual target (a red apple). As described in Pyykkö et al. [32], the task started with an “oh” sound and the presentation of an image of a red apple (5 visual angle) on the center of the screen. After the infant looked at the apple and 2,000 ms elapsed (or a maximum wait period of 4,000 ms elapsed), the apple was removed for 500 ms and subsequently reappeared in a randomly chosen location on the screen. Depending on an experimental condition, the apple reappeared either alone (one-object condition), together with four or eight distractors of one kind (e.g., four blue apples or four rectangle-shaped sliced apples, multiple-objects condition), or together with four or eight distractors of two kinds (e.g., two/four blue apples and two/four red sliced apples, conjunction condition). When the infant’s point of gaze hit the target or 4,000 ms elapsed from the start of the trial, the target made a spinning movement on the screen and a reward sound was played. There were four trials per condition in one session, i.e., 24 trials in total. A blank screen was presented for 500 ms between trials. In the anticipatory attention shifts task (adapted from [14]), we examined infant’s ability to anticipate the appearance of a visual stimulus in a predictable location [32]. The infants were presented first with an attention-getting stimulus (a pink pig face, 5 visual angle) in the center of the screen. When the infant’s gaze hit the central stimulus, the stimulus was removed and an auditory cue was presented together with two empty rectangles on both sides of the screen. A reward image (an animated duck) was subsequently shown in one of the two rectangles. The reward was presented on the same side (counterbalanced left or right) during the first eight trials (pre-switch). The side was then switched for the last eight trials (post-switch). There were a total of 16 pre-switch and 16 post-switch trials in two sessions. The time interval from the presentation of the two empty rectangles to the presentation of the rewards was contingent on the participant’s behavior. If the infant made a “correct” anticipatory saccade to the placeholder where the reward stimulus was about to appear, the reward was presented without a delay. If there was no correct anticipatory saccade, the reward was presented after a 1,000-ms delay. The attention to faces task was an overlap paradigm in which infant’s dwell time on a central stimulus (a non-face pattern or a face) was measured before its shift to a lateral distractor (following [29, 30, 46–48]). Each trial started with an attention-grabbing stimulus in the center of the screen. After a fixation at this stimulus, the attention-grabber was removed and two new stimuli were presented with a 1,000-ms onset asynchrony. First, a non-face pattern or a face on the center, then on the left or right side of the screen a black and white geometric shape superimposed by a cartoon. When the infant’s gaze moved to the lateral image or 1,000 ms elapsed, the cartoon picture turned into a video animation. The faces were pictures of two Black females with happy and fearful expressions. The non-face patterns were rectangular and phase-scrambled from the faces. Trials were presented in a random order and consisted of eight non-face trials and eight face trials (four happy and four fearful) per session (16 non-face trials and 16 face trial in total). Raw eye tracking data were preprocessed and analyzed offline by using a library of automated MATLAB (The MathWorks Inc.) functions [31]. The analyses followed the approach described in Pyykkö et al. [32] and no changes to the analyses of the eye tracking data were made for the current association analyses. The xy-coordinates corresponding to the two eyes were combined by taking a mean of the coordinates (or by using the eye with valid xy-coordinates if one of the coordinates for one of the eyes was invalid), extrapolated to fill missing data points (maximum of 200 ms), and median filtered with a moving window of nine samples to remove abrupt technical spike artefacts from the data. In each task, trials that failed to meet predetermined data quality criteria (i.e., violated upper limit of extrapolation) were excluded. Additional task-specific exclusion criteria were applied for the assessment of attentional dwell times in the face task so that trials with < 70% fixation on the central stimulus prior to attention shift and trials on which the shift occurred during a period of extrapolated data were excluded. From the visual search task, we extracted the visual search latency by calculating the mean latency of gaze shifts that entered the target area within a time period that started 150 ms after the onset of the target and ended 850 ms later. These limits are based on the convention that orienting responses shorter than 150 ms are considered anticipatory and orienting responses longer than 1,000 ms as delayed or missing responses. The search latencies were calculated by using data from the one-object condition because this condition was the only one that had a high rate of successful responses in all participants (mean 92%, range 20–100%) and because previous studies assessing infant’s processing speed as a predictor of cognitive development have used comparable measures [6, 7]. The latencies were calculated from a maximum of 8 trials per participant. Given variable number of successful search responses in the other (multi-object) conditions of the visual search task, the proportion of successful search responses (instead of latency) was calculated to obtain performance indicators for the multiple-objects and conjunction conditions. These indicators were calculated by counting the number of trials on which the point of gaze entered the target area within 2,000 ms and dividing the count by the total number of valid trials. From the anticipatory attention shifts task, we extracted the proportion of anticipatory gaze shifts to the correct side of the reward stimulus. As described in Pyykkö et al. [32], anticipatory responses were defined as entries of the point of gaze in the correct area of interest (i.e., side of the reward stimulus) within a 1,150-ms time window that started at the onset of the two empty rectangles and ended 150 ms after the onset of the reward stimulus. Again, the time window was extended to 150 ms after the onset of the reward stimulus on the basis of the commonly held assumption that gaze shifts that are shorter than 150 ms are typically considered anticipatory and could not, therefore, be reflecting a reactive saccade to the reward. Anticipatory gaze shifts were analyzed separately for the pre-switch and post-switch conditions. Trial numbers 1, 9, 17, and 25 were excluded from pre- and post-switch success rates as they were not predictable. Thus, for each condition, there was a maximum of 14 trials per participant. The data from the tasks assessing attention to faces was analyzed by computing the duration the infant gaze dwelled in the center area of interest (AOI) before an attention shift to the peripheral AOI occurred [32]. The analyses were censored at 3,500 ms meaning that if no attention shift occurred before this time-out value, the dwell time was 3500 ms. The dwell times < 150 ms were excluded. Dwell times were calculated separately for non-face patterns and faces. Both conditions had maximum of 16 trials by participant. For a participant to be included in the analyses of four visual attention measures, the participant needed to provide at least three valid trials for each condition of the task for a particular measure. For sensitivity analysis, as an overall visual attention score, each variable (excluding dwell time on faces, as its direction cannot necessary be defined as positive or negative) was ranked by quickness or success. Participant’s percentile on the seven conditions were summed. We assessed four domains of child development at 18 months of age. First, at the home visit child’s mother or caregiver was asked about the child’s word usage and socioemotional behavior. Later, at the clinic visit, the child was assessed on gross and fine motor development and executive functioning. Each assessment was performed once. Children were not asked for a re-visit if the assessment was unsuccessful. The development assessment tasks were previously used in the same location [49]. Child’s language development was assessed using a 100-word local language vocabulary checklist [50], which was translated into Chichewa and Chiyao and adapted for use in Malawi [49]. The mother or caregiver was presented with words from an existing list of words and asked to indicate whether the child has said the word. The child’s score was the cumulative total of positive responses (language score, maximum of 100). Child’s socioemotional development was assessed by using the Profile of Social and Emotional Development. The questionnaire was based on the Child Behaviour Questionnaire for Parents [51]. The mother or caregiver was asked 19 open-ended questions regarding problems in different domains of behavior (e.g., self-help, independence, attention, aggression, following rules, emotion control, playing with others, trouble in adjusting to changes), and the assessor rated answers as 0, 1 or 2 based on seriousness and frequency of the problem. Additionally, child’s maladaptive behavior was assessed by asking 11 open-ended questions regarding possible unusual or rare behaviors (e.g., showing no affection, unresponsiveness). The answers were scored as a binary variable where 0 indicate the absence of the behavior and 2 regular behavior. The sum score of the questionnaire across all 30 questions was reversed for higher score to indicate fewer socioemotional problems (socioemotional score, maximum of 60). Child’s motor development was assessed with the Kilifi Developmental Inventory tool developed in Kenya [52]. The assessment consisted of 36 gross motor (e.g., jumps with two feet, climbs onto platform; and an extra item of running) and 59 fine motor (e.g., kicks a ball, can do up button, puts coins in a box) items scored as binary variables (1 for successfully completed items). Some items were progressive, i.e., successful completion of the previous items was a requirement for the administration of the item (e.g., building tower of blocks was scored on a two-block interval, maximum of twelve blocks). A composite motor score was created by summing across all 95 items (max. score 95). Child’s mood, activity level, and interaction with the assessor were also rated. Mood was scaled from crying to laughing, activity level from unrousable to active, and interaction from avoidant to friendly. After the assessment of motor skills, the child’s executive function and working memory was assessed using a version of the A-not-B task [16]. Two cups were placed on a board in front of the child and the child saw a treat (a piece of corn puff) being put under either of the cups. The board with the two cups was subsequently hidden for 5 seconds and the child was distracted. When the board was returned, the child was asked to select the cup hiding the treat. The place of the treat was switched after two consecutive successful selections. The task was repeated ten times. Successful selections after a switch of a location were scored as 1 and summed (A-not-B score, maximum of 4). Only participants with the full ten attempts were included in analyses. Data on the child’s length, weight, head circumference, and mid-upper arm circumference were collected at enrollment, 9, and 18 months. Age- and sex-standardized anthropometric indices (length-for-age, LAZ; weight-for-age, WAZ; weight-for-length, WLZ; head circumference-for-age, HCAZ; and mid-upper arm circumference-for-age z-scores, MUACZ) were calculated using World Health Organization Child Growth Standards [53]. MUACZ was not available for enrollment measurements and thus, MUAC adjusted to age and sex was used instead when enrollment measurements were used. The change in anthropometric indices between 9 and 18 months of age (Δ9–18) was calculated as a measure of relative growth velocity adjusting for genetic potential which describes environmental factors affecting growth [54]. Mothers and family were assessed between the enrollment and 9 months on constructs of maternal cognition, maternal psychosocial well-being, socioeconomic status, and care practices. Details of these assessments are provided in Pyykkö et al. [32]. In the tests of maternal cognition, mothers were asked to complete tests assessing spatial cognition (mental rotation), working memory (digit span forward and backward test), and verbal fluency (listing foods and girls’ names). The tests of mental rotation consisted of five rows, one kind a figure in each row. Following an example figure, the mother was asked to point out rotated, not flipped figures. In the digit span tests, mothers were asked to repeat sequences of digits, first in the same order as presented and then in a reversed order. In both sets, the length of the sequence increased on every trial. In the listing tests, mothers were given 60 seconds to name as many foods as possible, then another 60 seconds to name as many girls’ names as possible. Maternal psychosocial well-being construct was assessed by questionnaires covering depression symptoms [55, 56], perceived stress [57], life events, and social support. Answer options were either yes/no or on different scales (numerical input from 0 to 2–4). Scores were summed within each subscale. Socioeconomic status was assessed on the basis of responses to questions concerning food insecurity in family (access to food, hunger) [58], living conditions (house material and equipment, water source), and the satisfaction of everyday needs (questions about money, food, laundry). Care practices consisted of an observation at home [59] and questions related to mother-infant bonding [60]. Observations and questions targeted mothers’ interaction styles and activities promoting cognitive, motor, and socioemotional development of the child. Mother-infant bonding consisted of questions about mother’s feelings toward the child, as described with different adjectives (a 4-point scale). Each variable was standardized and then summed within a domain to construct these four domains. Higher score indicates positive or better responses. We carried out the statistical analyses with Stata 15.1 (StataCorp) and R 4.0.0 (R Core Team). Graphics were produced with Microsoft PowerPoint (Microsoft Corporation) and R’s ggplot2 library [61]. We used correlation coefficients to examine associations between variables, and to allow for a clear, unified interpretation of the strengths of the associations between all comparisons. As some of the variables deviated from normal distribution, we used nonparametric methods: Spearman’s rank correlation coefficients for un-adjusted tests and Spearman’s partial rank correlation coefficients for adjusted tests (adjustments are listed inside parenthesis in the following chapters). We considered correlation coefficients > ∣0.20∣ significant following previous association analyses in infants. The sample size varied between analyses (n = 198–377) depending on the availability of valid data. For the constructs of the 9-month developmental scores, growth, and family characteristics, with the smallest sample size of a construct (n = 198, 254, and 262, respectively) and using a Bonferroni adjustment for the number of tests in each family of hypothesis tests (n = 16, 44, and 16, respectively), the two-sided p-values for the correlation coefficient ∣0.20∣ are < 0.08 (p = 0.076, 0.060, and 0.018, respectively). In the main analyses, we calculated correlation coefficients between measures of infant attention at 9 months and the developmental outcomes at 18 months. When necessary, the measure of interest was adjusted for related, but non-critical variability in task performance by using partial correlation tests (e.g., the proportion of successful visual searches in the context of interfering stimuli was adjusted for general proportion of successful visual searches in conditions that did not have the interfering elements). We focused on the four constructs that the eye tracking tasks were designed to measure at 9 months of age: (a) visual search latency using data from the one-object condition, (b) visual search interference (i.e., the proportion of successful visual searches using data from the conjunction condition; adjusted to the proportion of successful searches in one-object and multiple-objects conditions), (c) the ability to update anticipatory attention shifts after a change in stimulus contingency (i.e., proportion of correct anticipatory responses on post-switch trials; adjusted to proportion of correct anticipatory responses on pre-switch trials), (d) average dwell time for faces (adjusted to average dwell time for non-face patterns). The development outcomes at 18 months of age consisted of the four constructs measuring (e) language, (f) socioemotional behavior, (g) motor development (adjusted to child’s behavior during the test), and (h) A-not-B score. To obtain a variable reflecting the child’s behavior during the test, the mood, activity level and interaction with the assessor were ranked and combined to one variable, extracted from the first component of a principal component analysis. In secondary analyses, associations between growth and developmental outcomes were examined by linking (i) gestational age at birth, (j) anthropometric measurements (z-scores for length, weight, head circumference, mid-upper arm circumference) at 9 months of age (adjusted to the measurement at enrollment), and (k) change in anthropometric measurements between 9 and 18 months of age (adjusted to the measurement at 9 months of age) to the developmental outcomes at 18 months. In addition, we examined associations between family characteristics using (l) maternal cognition, (m) maternal psychosocial well-being, (n) socioeconomic status, and (o) care practices and the developmental outcomes at 18 months.