Introduction
Executive function (EF) and fluid cognitive skills are foundational aspects of neurocognition that rapidly develop across the first 5 years of life, are susceptible to early experience, and facilitate children’s ability to learn how to learn. Reference Zelazo, Blair and Willoughby1,Reference Ferrer, O’Hare and Bunge2 Policymakers, public health officials, pediatricians, and educators are keenly interested in promoting neurocognition in early childhood to prevent mental health disorders and to optimize educational and occupational outcomes. Reference Van Pelt, Lipow, Scott and Lowenthal3–Reference Haslam, Mejia, Thomson and Betancourt5 These efforts are of special interest in low and middle-income country (LMIC) contexts, where children face disproportional exposure to risk factors that undermine neurocognition, including infectious disease, environmental contaminants, and malnutrition. Reference Truter6–Reference Ssemata8
Although children’s stature (especially height for age) has long been used as a proxy of children’s neurocognition in LMIC contexts, Reference Sudfeld9 head circumference (HC) is a more proximal indicator of brain development. Reference Martini, Klausing, Luchters, Heim and Messing-Junger10,Reference Sandoval Karamian11 HC exhibits substantial growth in the first two years of life, which corresponds to normative increases in brain volume during this period. Reference Knickmeyer12 Identifying atypical patterns of HC growth in infancy and toddlerhood may facilitate early identification of children at risk for subsequent neurocognitive impairment.
Freire and colleagues Reference Freire13 conducted a systematic review of 115 studies that investigated the contemporaneous and prospective associations of HC with intelligence (IQ), academic, and occupational outcomes. Despite heterogeneity in geographic setting, research designs, and measurement of neurocognition, larger HC (including greater HC growth) was positively associated with IQ and academic outcomes in both general and at-risk (e.g., premature birth; low birth weight) populations. They highlighted the period from birth-2 years as a potential critical period, as changes during this interval were most strongly related to IQ. They also observed that many studies focused on extreme groups, such as children meeting criteria for microcephaly or macrocephaly. They also noted that the association between HC and IQ may be nonlinear; both microcephaly and macrocephaly were associated with lower IQ compared to normal HC. The authors recommended the use of repeated measurement of HC across the first two years of life as a screening approach, with special interest in LMIC contexts, and greater consideration of potential confounders.
Freire and colleagues did not present results separately for studies by geographic region, and we are unaware of any reviews of HC and neurocognition specific to LMICs. Nonetheless, individual studies generally support Freire’s conclusions. For example, three studies have reported on the Etiology, Risk Factors and Interactions of Enteric Infections and Malnutrition and the Consequences for Child Health and Development (MAL-ED; n = 1210) cohort. In an initial analysis of the full sample, Scharf and colleagues Reference Scharf14 reported that HC was associated with improved cognitive development (using the Bayley scale) at age 2. In a longitudinal follow-up of the South Indian subsample of MAL-ED (n = 235), Koshy and colleagues Reference Koshy15 reported that HC < -3 SD, using World Health Organization standards, was associated with poorer cognitive development (using the Wechsler Preschool and Primary Scale of Intelligence) at age 5 after adjustment for family, maternal, and child covariates. Conversely, in an updated analysis of the full sample, Nicolau and colleagues Reference Nicolaou16 reported no association between HC and cognitive, gross, motor, or language skills at 6, 15, and 24 months of age, which they attributed to a more expansive set of covariates.
Two recent studies investigated the association between HC and cognitive development (using the Mullen scales) in a Guatemalan cohort. Connery and colleagues Reference Connery17 reported significant associations between a range of HC values and Mullen scores obtained 11 months later. Lamb and colleagues Reference Lamb18 extended that work by considering developmental trajectories of HC growth. They concluded that continuous measures had stronger associations than cutoff scores (i.e., stunting, microcephaly) and that HC was a more robust predictor than traditional measures of linear growth.
Do and colleagues conducted a prospective cohort study of preterm infants in Vietnam. Reference Do19 Although most infants demonstrated catch-up in weight and length (relative to WHO norms), their rates of head growth were delayed. Infants who exhibited slower improvements in HC from 3–12 months were at increased risk for cognitive problems (using the Bayley scale) at 24 months.
This study addresses three gaps in the literature. First, whereas most studies have focused on the association between HC and cognition in infancy and toddlerhood, we extend this work to the early childhood period. Moreover, where many previous studies focused on general measures of cognition, we focus on EF and fluid cognitive processes that contribute to early learning and self-regulation. Second, we focus on trajectories of continuous measures of HC, rather than time-specific indicators measures of HC (including thresholds for microcephaly) or simple difference scores in HC, as our focal predictor. We test whether inter-individual differences in intra-individual change in HC across the first two years of life uniquely predict neurodevelopment at age 4, above and beyond parental and household covariates that influence neurocognition. Third, we explore sensitive periods. Specifically, we consider the minimal span of time, within the first two years of life, for which changes in HC are predictive of neurocognition at age four years.
Methods
Study procedures
The Linda Kizazi Study was established in 2018 to study how maternal HIV infection affects mother-to-child viral transmission. The study is described in detail elsewhere. Reference Begnel20 Equal numbers of women living with and without HIV were enrolled in pregnancy and followed with their children through 4 years of age. Enrollment criteria for all women were: 28–42 weeks pregnant, 18–40 years old, planned to breastfeed, did not have a planned Cesarean section, were not diagnosed with a serious medical condition, and if living with HIV had received ≥6 months of antiretroviral therapy. Detailed demographics and medical history were collected at enrollment, and participants and their infants were evaluated clinically with anthropometry at delivery, weeks 6 and 10, then every 3 months until 24 months of age. To facilitate clarity, we refer to week 6 as “month 1” and week 10 as “month 3” throughout. A second project grant was awarded in 2022 that included cognitive assessments, including of EF and fluid cognitive skills, when children were four-years-old. The current study makes joint use of early life HC and age 4 cognitive data.
Cognitive assessments were completed by four local child examiners who were fluent in English and Kiswahili. All assessments were administered in Kiswahili or, in a few cases, a combination of English and Kiswahili based on the child’s language preference. We used a previous Kiswahili translation of the KABC-II, and in-country study team members provided an updated translation (and back translation) of the EF Touch assessment. Prior to completing study assessments, examiners received didactic training in standardized assessment and administration of the EF-Touch and KABC-II assessments (described below). Prior to testing study participants, examiners completed at least 10 video-recorded practice assessments with children in the target age range. These practice administrations were reviewed by one of the investigators (BC) to ensure ≥ 90% scoring agreement and standardized administration without significant errors affecting validity. Throughout the study, 10% of the administrations were video recorded for quality assurance.
Measures
Anthropometric measurements
At each visit, study clinicians trained in child anthropometry took physical measurements of the child, including HC (cm), weight (kg), length/height (cm), and middle-upper arm circumference (cm). Measurements were taken twice and the average of both values for each measure was used for analyses.
Executive function touch in tangerine
Executive function (EF) skills were measured using Tangerine™ EF Touch, a tablet-based battery of tasks that have been previously used with preschool-aged children in Kenya. Reference Willoughby, Piper, Kwayumba and McCune21,Reference Willoughby, Piper, Oyanga and Merseth King22 The battery includes two tasks (Training, Bubbles) that orient children to the experience of touching the screen to make responses. EF skills were indexed based on performance on up to four EF tasks. Two tasks – Animal Go/No-Go (a go/no-go task) and Silly Sounds (a Stroop-like task) – measured inhibitory control. These tasks required children to respond within 3000–3500 milliseconds to items that involve overcoming a prepotent response. Two additional tasks – Something’s the Same (an item selection task) and Pick the Picture (a self-ordered pointing task) – measured cognitive flexibility and working memory, respectively. Children must first demonstrate their understanding of task rules by successfully completing training items in order to advance to test items. Children are given two attempts to successfully complete training items; otherwise, the task is discontinued. Overall, rates of administration were high across tasks: 84% of children passed Animal Go/No-Go, 80% passed Silly Sounds, 94% passed Something’s the Same, and 92% passed Pick the Picture training items. Children were most challenged by the Silly Sounds training; in this task, children hear a dog “woof” or a cat “meow” and must quickly select the animal that does not make that sound. Tasks assessing cognitive flexibility and working memory (Something’s the Same; Pick the Picture) had higher rates of administration than inhibitory control tasks.
Following precedent, Reference Camerota, Willoughby and Blair23 we created an EF composite score that reflected average accuracy across completed tasks. For over two thirds of participating children (69%), the EF composite score incorporated task data from all four EF tasks; 16% were based on three EF tasks, 9% on two EF tasks, and only 6% of EF composite scores were based on one EF task. Elaborated task descriptions are provided in Supplemental Materials.
Kaufman assessment battery for children, second edition (KABC-II)
The KABC-II is a “culturally fair” assessment of cognitive ability that is widely used across various contexts due to the minimal extent to which cultural and ethnic differences are associated with assessment scores. Eight subtests were administered (Atlantis, Conceptual Reasoning, Face Recognition, Number Recall, Rebus, Triangles, Word Order, Hand Movements). These subtests are part of the KABC-II learning, simultaneous/visual processing, and sequential/short-term memory scales. Seven of these subtests (all but Hand Movements) were combined to form the Mental Processing Index (MPI) score, which indexes fluid cognitive skills.
Analytic strategy
We used latent growth curve models (LCMs), a subset of structural equation models, to address all study aims. Reference Bollen and Curran24 LCMs were ideal for our study because they permit the simultaneous estimation of change in HC and the regression of distal outcomes (i.e., EF Touch composite; KABC-II MPI) onto growth parameters. Analyses proceeded in four steps. First, in results that are not presented, we estimated a series of unconditional growth models, including linear, quadratic, and unstructured models, to establish the functional form of change in HC from 1–24 months. An unstructured model provided the best fit to the data. The unstructured model was parameterized by fixing the factor loadings of the first and last timepoints to values of 0 and 1, respectively, and freely estimating the remaining loadings. As a result, the intercept factor reflected individual differences in HC at 1 month of age, and the slope factor reflected individual differences in total change in HC from 1–24 months. Second, we estimated a series of conditional LCMs, in which neurocognitive outcomes at age 4 were separately regressed on the intercept and slope (total change) parameters and covariates including child age at follow-up, child sex, and maternal education. Following McCormick, Reference McCormick, Curran and Hancock25 the slope term was regressed onto the intercept factor, which improved the interpretability of results (see Supplemental Fig 1 for an exemplary path diagram). Specifically, in typical LCMs with distal outcomes, the slope and intercept factor coefficients both capture some portion of shared variance with the distal outcome. By regressing the slope factor onto the intercept factor, the slope factor represents the unique association of HC growth on the distal outcome, which was our primary research question. Third, we conducted a series of robustness tests to establish the consistency of predictive associations between growth in HC and neurocognitive outcomes. Specifically, we iteratively re-estimated conditional LCMs while systematically excluding children with prenatal exposure to HIV and/or who experienced atypical growth at any point in the first 24 months, including wasting [defined as having a weight-for-length z-score (WLZ) of less than -2], being underweight [defined as having a weight-for-age z-score (WAZ) of less than -2], having extreme HC [defined as having a HC-for-age z-score (HCAZ) outside of |3|] or extreme WLZ (defined as having a WLZ outside of |3|). We chose these metrics because they are routinely used in the literature to index children’s health status. We used forest plots to characterize the robustness of results across subsamples. Fourth, to address questions about sensitive periods, we iteratively re-estimated conditional LCMs by reducing the age period during which changes in HC were considered (e.g., 1–24 months, 1–21 months, etc.). We used forest plots to characterize changes in the effect of the slope term on neurocognitive outcomes across progressively shorter intervals of time.
All models were estimated using MPlus software Reference Muthén and Muthén26 with a full information maximum likelihood estimator that accommodated missing HC data (see Supplementary Fig 2 for an example MPlus script for the conditional LCM). Model fit was evaluated using a chi square test of absolute fit and model fit indices, including the Root Mean Square Error of Approximation (RMSEA) and Comparative Fit Index (CFI). RMSEA values < 0.05 and 0.05–0.09 are indicative of excellent and acceptable fit, respectively. CFI values > 0.95 and of 0.90–0.95 are indicative of excellent and reasonable/acceptable fit, respectively.
Results
Participant characteristics
To be maximally inclusive, this analysis included 182 children who were enrolled in the Linda Kizazi Study and who had any of the variables used in our analysis (i.e., demographic covariates, HC, and/or neurocognitive outcomes). HC was measured up to nine times across the first 24 months of life. In total, there were 1,012 unique person-observations (Mn = 5.56, Range = 0–8 measurements/child).
Participant characteristics are presented in Table 1. Nearly half (48%) of participating mothers had a primary school education level or less and 46% of children were HIV-exposed but uninfected (born to women living with HIV). Children were approximately equally distributed by sex (55% male). At the 48-month visit, 105 children completed at least one neurocognitive assessment; 103 children completed the KABC-II and 99 children completed the EF Touch. Children who completed neurocognitive assessments contributed more HC measurements (M = 6.44 vs. M = 4.36; p < 0.01) and were more likely to have been categorized as having experienced wasting (36% vs. 19%; p = 0.03) at some point between 1–24 months than children who did not complete any child neurocognitive assessment. There were no significant differences in HIV exposure status, child sex, maternal education level, or other growth metrics between children who completed the neurocognitive assessments and those that did not (see Table 1).
Table 1. Sample description
Note: Prob denotes p-values for contrasts between participants with and without observed neurocognitive outcome data. Never experienced wasting is defined as never having a weight-for-length z-score (WLZ) less than -2 from 1–24 months; Never experienced underweight is defined as never having a weight-for-age z-score (WAZ) less than -2 from 1–24 months; Within 3 Zs of WLZ is defined as always having a WLZ between -3 and 3 from months 1 to 24; Within 3 Zs of HCAZ is defined as always having a head circumference-for-age (HCAZ) z-score between -3 and 3 from months 1 to 24; all z scores refer to World Health Organization values by sex and age.
Developmental changes in head circumference
Descriptive statistics for repeated measures of HC from 1–24 months are summarized in Table 2. The observed pattern of means indicates a nonlinear pattern of growth, with larger increases occurring in the first versus second year of life. The rank order of individual differences in HC was strong (rs = 0.28–0.72), especially for time adjacent measures. However, bivariate correlations between time-specific measures of HC with EF (rs = −0.04–0.20) and KABC fluid cognitive processes (rs = −0.04–0.28) were of small to moderate magnitude and (with two exceptions) not statistically significant. A spaghetti plot of observed changes in HC from 1–24 months is depicted in Figure 1. This plot highlights variation in initial and ending levels of HC, as well as additional evidence for nonlinear changes in HC over time.
Figure 1. Children exhibited significant variability in their initial and final HC, as well as in their rate and total change in HC from 1 to 24 months.
Table 2. Descriptive statistics and bivariate correlations of head circumference and neurocognitive outcomes
Note. Total N = 182; * p < 0.05; ** p < 0.01; *** p < 0.001; HC, head circumference; EF, executive function composite; MPI, KABC-II Mental Processing Index.
We estimated an unstructured LCM to characterize nonlinear changes in HC across 1–24 months, which fit the data well (χ2 [df = 33] = 46.41, p = 0.06; RMSEA = 0.048; CFI = 0.97). The factor loading estimates (λ) are interpreted as the average proportion of total change that occurred by a given assessment period. Results demonstrated nonlinearity in growth of HC: λ = 0, 0.16, 0.54, 0.72, 0.79, 0.85, 0.92, 0.98, and 1.0 for HC measures at 1, 3, 6, 9, 12, 15, 18, 21, and 24 months of age, respectively (e.g., the estimated loading of 0.85 for the 15-month measure indicates that 85% of the total change in HC from 1–24 months occurred by age 15 months). The variances of the intercept and slope factors were statistically significant (both p < .001), which indicated meaningful variation between children’s initial HC at 1 month of age and in the total amount of change in HC from ages 1–24 months. The intercept and slope terms were negatively correlated, φ = −0.37, p < 0.01. Hence, children with larger HC at age 1 month exhibited less total growth in HC from ages 1–24 months.
Conditional growth model
We estimated a series of conditional LCM models, in which the EF composite and KABC-II MPI scores were separately regressed on HC intercept and slope factors, as well as child age and sex and maternal education. As summarized in Table 3, children who exhibited the greatest improvement in HC performed better on the EF composite (β = 0.33, p = 0.02; model 1) and KABC-II MPI (β = 0.35, p < 0.01; model 7).
Table 3. Contributions of initial (intercept) and total change (slope) in HC (1–24 months) on neurocognition (48 months)
Note: * p<0.05; ** p<0.01; *** p < 0.001; DF, degrees of freedom; CI, confidence interval; Never experienced wasting is defined as never having a weight-for-length z-score (WLZ) less than -2 from 1-24 months; Never experienced underweight is defined as never having a weight-for-age z-score (WAZ) less than -2 from 1-24 months; Within 3 Zs of WLZ is defined as always having a WLZ between -3 and 3 from months 1 to 24; Within 3 Zs of HCAZ is defined as always having a head circumference-for-age (HCAZ) z-score between -3 and 3 from months 1 to 24; all z scores refer to World Health Organization values by sex and age.
Robustness testing
We iteratively re-estimated the above models for subsamples of children who from 1–24 months were (1) HIV-unexposed, (2) never experienced wasting, (3) were never underweight, (4) had WLZ within -3 and 3, and (5) had HCAZ within −3 and 3. The point estimates for the association between total change in HC and the EF composite was highly consistent across subpopulations (Table 3; models 2–6). The same was true for the association between total change in HC and the KABC-II MPI (Table 3; models 8–12).
Critical periods
Finally, we re-estimated the above models while successively restricting the period in which HC change was considered (e.g., 1–24, 1–21, 1–18 months, etc.). Point estimates were highly similar for periods of HC change that spanned 1–15, 1–18, 1–21, and 1–24 months, suggesting that the predictive associations between total change in HC and the EF composite (Figure 2) and KABC-II MPI (Figure 3) were evident within the first 15 months of life. This result was consistent in both the full and subgroup samples.
Figure 2. Markers represent the standardized regression coefficient of growth in HC on EF Touch composite score; horizontal lines reflect 95% confidence intervals of the coefficient. Children with the largest gains in HC in the first 24 months of life exhibit better executive function skills at 48 months of age. This association is evident as early as the first 15 months of life. These associations are largely robust when excluding children who were HIV exposed or exhibited atypical growth at any point between months 1 and 24, including ever experienced wasting (WLZ < −2), being underweight (WAZ < −2), extreme HCAZ (<−3 or >3) or extreme WLZ (<−3 or >3).
Figure 3. Markers represent the standardized regression coefficient of growth in HC on the KABC-II Mental Processing Index; horizontal lines reflect 95% confidence intervals of the coefficient. Children with the largest gains in HC in the first 24 months of life exhibit better mental processing skills at 48 months of age. This association is evident as early as the first 15 months of life. These associations are largely robust when excluding children who were HIV exposed or exhibited atypical growth at any point between months 1 and 24, including ever experienced wasting (WLZ < −2), being underweight (WAZ < −2), extreme HCAZ (<−3 or >3) or extreme WLZ (<−3 or >3).
Discussion
In this cohort, children who exhibited the greatest gains in HC across the first two years of life exhibited comparatively better EF and fluid cognitive skills at age four. This pattern persisted after adjusting for child age, child sex, and maternal education, and was largely robust across subsamples of children who exhibited atypical physical growth or exposure to HIV. To our knowledge, this is the first study that has explored these associations in Kenya.
The most innovative feature of this study was our consideration of developmental trajectories of HC, rather than time-specific indicators or simple difference scores, as the focal predictor of neurocognitive outcomes. Latent growth curve models provide a flexible approach for summarizing inter-individual differences in intra-individual change, while accounting for unbalanced or missing data and varied functional forms of change. Unlike other studies, we parameterized models in ways that prioritized change in HC, independent of the impact of initial level of HC, as our primary predictor. Reference Muthén and Muthén26 Our results align with a recent a prospective study of preterm infants in Vietnam, in which infants who exhibited slower rates of HC growth from 3–12 months were at increased risk for mental delays at 24 months. Reference Do19
Our focus on developmental trajectories of HC also facilitated a straightforward approach for testing questions about critical periods that Freire et al. raised in their systematic review. We observed nonlinear patterns of change in HC across the first 24 months of life, which is consistent with normative patterns of structural brain development that occur during this period of time. Reference Knickmeyer12 We demonstrated that individual differences in HC growth from birth through 15 months were predictive of EF and fluid cognitive skills at age 4, with similar precision to what was observed from birth through 24 months. If replicated, these results highlight the potential of using individual trajectories of HC during infancy as a low-cost method for the early identification of children who may benefit from more in-depth assessment and/or early interventions (e.g., nurturing care, dietary intervention, and other general health promotion practices). The consideration of continuous measures of HC (including trajectories of HC) may help to identify a broader number of children than traditional cutoffs, including stunting and microcephaly. Reference Lamb18
Most previous studies have focused on the impact of HC on general measures of cognition in infancy and toddlerhood. We extended those studies by focusing on EF and fluid cognitive skills in early childhood. This is notable because early childhood is a period during which EF and fluid cognitive skills undergo rapid change and facilitate children’s early learning and self-regulation. Reference Zelazo, Blair and Willoughby27
This cohort has several strengths and a few important limitations to note. Strengths include the long-term follow-up from birth, close clinical monitoring of the children, and comprehensive assessment of multiple domains of functioning. Because participants were originally recruited to study mother-to-child viral transmission among children with prenatal exposure to HIV, the study design minimized cohort heterogeneity by household, demographic and other health exposures. Participants were recruited from a single community clinic and provided with exceptionally high-quality health care, including optimized antiretroviral therapy initiated before or in early pregnancy among the mothers living with HIV. This approach limits the generalizability of our results to other populations in LMICs but has the benefit of reducing bias from measured and unmeasured sociodemographic and HIV-associated confounders.
Only 58% of children enrolled who were enrolled in the original birth cohort participated in the neurocognitive assessment at age 4. Although there were few differences between children who did and did not participate in the age 4 assessment, we cannot rule out selective attrition (e.g., for unmeasured variables). Finally, we addressed a focal question regarding the association between early trajectories of HC and neurocognition. Given the small sample size, we did not address broader questions about how early life experiences – including variations in nutrition, infection, and nurturing care – may affect both trajectories of HC and neurocognition and/or moderate these associations. These questions should be addressed in future studies.
In sum, normative improvements across the first two years of life – and as early as the first 15 months of life – were predictive of EF and fluid cognitive skills at age four in this cohort of Kenyan children. Though we acknowledge that changes in HC are a crude proxy for structural changes in brain development, this measure is inexpensive to obtain, easily incorporated into child wellness and immunization visits, and may be useful in identifying children who would benefit from more in-depth neurocognitive assessments and early intervention activities.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S2040174425100275.
Acknowledgments
None.
Financial support
This study was funded by the Eunice Kennedy Shriver National Institute of Child Health and Human Development (P01HD107669, R01HD023412, R01HD092311), and the Canadian Institutes of Health Research (390237). Additional research support was provided by the Global Center for Integrated Health of Women, Adolescents and Children (Global WACh), the UW Kenya Research and Training Program (KRTC), and the UW Center for AIDS Research (P30 AI027757, National Institute of Allergy and Infectious Diseases). The views expressed in this manuscript are those of the authors, and they do not necessarily represent the opinions and positions of the funders.
Competing interests
None.
Ethical standards
The authors assert that all procedures contributing to this work comply with the ethical standards of the National Guidelines for Ethical Conduct of Biomedical Research Involving Human Participants in Kenya and with the Helsinki Declaration of 1975, as revised in 2008, and has been approved by the Kenyatta National Hospital-University of Nairobi Ethics and Research Committee (P472/07/2018) and the University of Washington Institutional Review Board (STUDY00004006) Primary caregivers provided active consent for their children to participate.
Open access
