Early Child Development Trajectories and Adult Cognitive Function: A 50-Year Longitudinal Study in Guatemala
Global Context and Life-Course Cognitive Development in LMICs
Prevalence of Developmental Risk:
In low-income and middle-income countries (LMICs), approximately of children younger than of age were estimated in 2010 to be at risk of poor cognitive and physical development when using stunting and extreme poverty as proxy indicators.
The majority of children at risk globally reside within LMICs, making these populations the primary global targets for developmental and nutritional interventions.
Significance of Early Childhood Cognitive Potential:
Early childhood development and intellectual functioning serve as critical determinants of survival, long-term health status, academic success, and human capital accumulation across the lifespan.
Ensuring that all children reach their full developmental potential represents a core priority of the international sustainable development agenda.
Gaps in Life-Course Evidence:
Despite widespread calls for early interventions, tracking intellectual functioning across time has been hindered by the complex etiology of child development and the methodological challenges of assessing cognitive function across different life stages in diverse populations.
The first years of life constitute the most rapid and vulnerable period of structural brain growth, cellular differentiation, and neural connectivity.
Most evidence examining cognitive developmental trajectories across the entire life span originates from high-income countries, leaving substantial uncertainty regarding how early cognitive performance tracks into adulthood within low-resource, transitioning environments.
Study Design and the INCAP Longitudinal Cohort
Original Intervention Framework:
Data originate from the Institute of Nutrition of Central America and Panama (INCAP) Nutrition Supplementation Trial Longitudinal Study, conducted in four rural Guatemalan villages.
The original trial was designed to test the prospective hypothesis that providing protein-energy nutritional supplementation during the preschool years prevents growth failure and improves childhood mental development.
Nutritional Interventions and Implementation (1969–1977):
Villages were assigned pairwise to receive one of two nutritional supplements:
Atole: A high-protein energy gruel providing per serving, containing vegetable protein and micronutrients.
Fresco: A protein-free, low-calorie sweetened fruit drink providing per serving.
Both supplements were made available daily at centrally located feeding stations, offered twice per day to all community members from 1969 through 1977.
Precise individual supplement intakes were measured and recorded for pregnant and lactating women, as well as children under of age.
Longitudinal Cohort Recruitment and Follow-Up Waves:
A total of children born between 1962 and 1977 were enrolled in the original investigation, either because they were under of age at study inception or were born in the villages during the trial implementation.
Long-term multi-wave follow-up assessments of the cohort were subsequently conducted in:
1988–1989 (adolescence and young adulthood; ages )
2002–2004 (established adulthood; ages )
2015–2017 (mature adulthood and midlife; ages )
Community-level engagement was maintained across the decades through periodic study feedback delivered directly to participating villages.
Analytical Sample Selection and Follow-Up Flow
Inclusion Criteria for Trajectory Modeling:
Modeling childhood cognitive trajectories required participants to have at least valid cognitive measurements out of the possible testing time points between and of age.
This conservative threshold ensures robust modeling of individual latent slopes and intercepts while retaining statistical power.
Inclusion Criteria for Adult Cognitive Associations:
Participants in the trajectory sample were included in the adult outcome analyses if they possessed at least one valid measurement of Raven's Progressive Matrices (RPM) or InterAmerican Reading Series literacy administered after of age.
Those whose last cognitive test occurred prior to age () were excluded from adulthood analyses.
Sample Flow Breakdown:
Original cohort born in study villages (1962–1977):
Completed at least one childhood cognition measurement:
Completed at least three childhood cognition measurements:
Completed at least one lifetime Raven assessment:
Completed at least three childhood measurements and one Raven assessment:
Excluded due to no Raven measurement after of age:
Final sample with trajectories and adult Raven measurements:
Final sample with trajectories and adult literacy measurements:

Attrition Analysis:
The final adult analytical sample () demonstrated baseline socioeconomic, parental, and physical characteristics comparable to participants lost to adult follow-up ().
The only significant difference observed was a higher proportion of men lost to follow-up over the 50-year interval, primarily driven by out-migration.
Cognitive Function Instruments and Assessment Protocols
Infant Battery (, , and of age):
Administered within an exact time window surrounding the infant's , , and milestones.
Developed specifically for rural Guatemala by expert psychologists and psychiatrists as a composite battery drawn from standard validated scales:
Bayley Scales of Infant Development
Gesell and Amatruda’s Developmental Diagnosis
Cattell Infant's Intelligence Scales
Merrill-Palmer Standards of Physical and Mental Growth
All items were translated into Spanish and culturally adapted.
The final Composite Infant Score (CIS) aggregates performance across discrete developmental items assessing sensory, motor, and mental development.
Preschool Battery (, , , , and of age):
Administered within a window of each child's respective birthday.
Age-graded testing structure:
Ages and : A shortened battery of tests.
Ages , , and : A comprehensive battery of tests.
Core cognitive domains evaluated included verbal comprehension, general knowledge, quantitative ability, and memory capacity.
Prior structural validation via factor analysis established two robust, predictive factors: a General Factor and a Memory Factor.
Adult Non-Verbal Intelligence: Raven's Progressive Matrices (RPM):
Administered across three adult rounds: 1988–1989 (ages ), 2004–2005 (ages ), and 2015–2017 (ages ).
Evaluates non-verbal abstract reasoning and pattern analysis using visual matrix problems with increasing difficulty.
Protocol adaptation: Standard RPM utilizes five sets of items ( items total); however, local pilot testing showed floor effects beyond the third set, resulting in the standard administration of the first three sets ( items total).
Total score is calculated as the sum of correct answers ().
Test properties: Demonstrates high internal consistency, construct validity, and excellent test-retest reliability () within this Guatemalan cohort.
Adult Literacy and Reading Comprehension: InterAmerican Reading Series:
Administered during the 2004–2005 and 2015–2017 follow-up rounds (mean participant age at last assessment: ).
Screening procedure: Participants without completed primary education or unable to read a sample sentence from the local newspaper completed a screening pretest ().
Full administration: Participants passing the screen (answering all items correctly) or having completed secondary school were assigned a base score of , which was then summed with their performance on the formal Serie Interamericana reading and writing composite modules.
Validated with robust test-retest reliability and internal consistency in this population.
Statistical Methodology and Analytical Framework
Standardization of Childhood Scores:
To enable valid comparative modeling across disparate assessment tools, raw scores on the CIS (, , and ) and preschool composite batteries (, , , , and ) were converted to standardized -scores adjusted for age relative to the internal cohort population.
Latent Class Growth Analysis (LCGA):
Utilized to cluster participants () into distinct developmental trajectories spanning to ().
LCGA models distinct latent classes defined by unique intercepts and growth slopes across repeated measurements, capturing population heterogeneity.
Models were adjusted for participant sex and clustered by maternal identification number to account for sibling non-independence.
Class selection was determined empirically using Bayesian Information Criteria, Akaike Information Criteria, entropy, class size proportions, and qualitative developmental plausibility using M-Plus (Version 7.0).
A sensitivity analysis restricted to participants having at least one infant score and at least one preschool score () was executed to evaluate selection bias from birth-year design effects.
Predictor Analysis Using Ordinal Logistic Regression:
Trajectory classes were treated as an ordered categorical dependent variable (, , ).
Proportional odds cumulative logit models evaluated family- and community-level determinants:
Household socioeconomic status (SES) score
Maternal and paternal years of completed schooling
Maternal and paternal height (, representing intergenerational human capital)
Household family size (number of cohabiting residents)
Distance of domicile to the village nutritional supplementation centre ()
Village size (community level)
Adjusted models included village of residence (supplement type) and exposure to Atole supplementation during the first of life.
Missing covariate data were handled using multiple imputation in SAS Version 9.4 (PROC MI generating imputed datasets via fully conditional specification, pooled with PROC MIANALYZE).
Modeling Trajectory Associations with Adult Function:
Mixed-effects linear models (PROC MIXED) estimated the intercept across all repeated adult RPM non-verbal intelligence assessments, adjusting for age at the time of each adult measurement.
Analysis of variance (ANOVA) evaluated associations between childhood trajectory membership and the final available adult literacy score.
Parallel models assessed adult performance stratified by tertiles of the single cognitive measurement collected at age () to benchmark trajectory predictive utility against cross-sectional evaluations.
Childhood Cognitive Development Trajectories
Class Structure and Membership:
Latent Class Growth Analysis identified three distinct longitudinal cognitive trajectories from age to ():
Low Trajectory: ( of sample)
Average Trajectory: ( of sample)
High Trajectory: ( of sample)
Developmental Pathways Across Age Milestones:
Timing of Divergence: Statistically significant developmental divergence between the trajectories emerges as early as of age and expands markedly across the first of life.
Low Trajectory Pattern:
Starts at with a mean -score of .
Exhibits a sustained downward developmental velocity through infancy: at , at , and at .
Reaches a nadir at () with .
Stabilizes slightly during early primary schooling to at and at .
Average Trajectory Pattern:
Maintains relative developmental stability across the entire observation period.
Trajectory coordinates: at , at , at , at , at , at , at , and at .
High Trajectory Pattern:
Starts slightly above the population average at ().
Demonstrates rapid developmental acceleration: at , at , at , and at .
Continues expanding upward during the transition into schooling: at , at , and peaks at at .

Trajectory Parallelism and Schooling:
While trajectories diverge sharply from birth to age , they track in an essentially parallel orientation from to .
This parallel configuration indicates that relative developmental ranking remains fixed once children transition into school age.
Sociodemographic and Physical Predictors of Trajectories
Baseline Household and Parental Characteristics by Trajectory:
Community Scale: Proportion residing in large villages did not differ by trajectory: Low (), Average (), High (), with .
Physical Accessibility: Distance to the feeding center showed no significant variation across groups: Low (), Average (), High (), with .
Household Socioeconomic Status (SES Score): Highly stratified across groups: Low (), Average (), High (), with .
Maternal Formal Schooling: Low (), Average (), High (), with .
Paternal Formal Schooling: Low (), Average (), High (), with .
Maternal Height: Low (), Average (), High (), with .
Paternal Height: Low (), Average (), High (), with .
Family Size: Household member count was inversely graded but not statistically significant in crude bivariate testing: Low (), Average (), High (), with .
Participant Sex: Gender distribution was balanced: Low ( male), Average ( male), High ( male), with .
Child Physical Growth at :
Height at : Markedly divergent across trajectory classes: Low (), Average (), High (), with .
Weight at : Low (), Average (), High (), with .
Proportional Odds Logistic Regression Models ():
Predictor Variable | Unadjusted Proportional Log Odds () | Adjusted Proportional Log Odds ()* |
|---|---|---|
Socioeconomic Status (SES) Score | () | () |
Family Size (Number of people) | () | () |
Distance to Supplementation Centre () | () | () |
Maternal Schooling () | () | () |
Paternal Schooling () | () | () |
Maternal Height () | ( ) | () |
Paternal Height () | () | () |
*Note: Adjusted models control for all other listed sociodemographic predictors and early nutritional supplementation with Atole during the first of life. Sex and age were controlled during LCGA trajectory formation.
Multivariable Predictor Takeaways:
Higher household SES, greater maternal education, and greater maternal height remain significant independent protective factors promoting membership in higher developmental trajectories.
Larger family size exhibits a significant independent negative association (), reflecting resource dilution in larger households.
Paternal schooling and paternal stature attenuate toward null associations after controlling for maternal capital and household socioeconomic resources.
Correlation Patterns Between Childhood Assessments and Adult Cognition
Pairwise Correlation Architecture ( with at least cognitive measures):
Pairwise Spearman correlation coefficients illustrate how internal test validity and external predictive capability evolve from infancy to mature adulthood.
Infant Battery Performance (, , and ):
Within-battery infant correlations:
vs. : ()
vs. : ()
vs. : ()
Correlations with adult RPM scores (ages ) remain universally weak:
Infant scores at correlate () at ages , () at ages , and () at ages .
Infant scores at correlate () at ages , () at ages , and () at ages .
Infant scores at correlate () at ages , () at ages , and () at ages .
Preschool Battery Performance ( to ):
Predictive power increases steadily with age:
Age correlates weakly with adult RPM: (), (), and ().
Age shows modest tracking: (), (), and ().
Age achieves moderate stability: (), (), and ().
Age yields the strongest single-age correlation with adult RPM: (, ), (, ), and (, ).
Age correlates strongly: (, ), (, ), and (, ).
Correlations Involving Trajectories and Adulthood Stability:
Overall trajectory classification (coded , , ) correlates with adult RPM at (, ), (, ), and (, ).
Trajectory classification correlates heavily with later childhood tests: at age , at age , and at age .
Within-adulthood non-verbal intelligence displays extreme stability across middle and late adulthood:
RPM at vs. : ()
RPM at vs. : ()
RPM at vs. : ()
Life-Course Cognitive Tracking and Adult Cognitive Outcomes
Adult Non-Verbal Intelligence Outcomes (Raven's Progressive Matrices):
Across the cohort, the overall individual adult RPM score mean and standard deviation were .
Mixed models estimating the random intercept of repeated RPM measures demonstrated clear, statistically significant separation across trajectory groups ():
Low Childhood Trajectory ( total analytical sample): Mean intercept =
Average Childhood Trajectory: Mean intercept =
High Childhood Trajectory: Mean intercept =
The absolute disparity between the Low and High trajectory groups is , representing a gap of based on the cohort mean.
Stratification by single cognitive tertiles () showed an identical graded pattern ():
Low Tertile:
Medium Tertile:
High Tertile:
Difference between extreme tertiles: .
Adult Literacy and Reading Comprehension Outcomes (InterAmerican Series):
Across the sample, the overall adult literacy score mean and standard deviation were .
Final literacy scores at a mean age of showed robust, highly significant differences by trajectory ():
Low Childhood Trajectory ( total analytical sample): Mean score =
Average Childhood Trajectory: Mean score =
High Childhood Trajectory: Mean score =
The absolute disparity between Low and High childhood trajectory groups is , equal to of the cohort mean.
Stratification by single cognitive tertiles () similarly demonstrated strong separation ():
Low Tertile:
Medium Tertile:
High Tertile:
Difference between extreme tertiles: .
Methodological Strengths, Limitations, and Policy Implications
Key Methodological Strengths:
Spans of prospective follow-up in a single well-characterized birth cohort in a transitioning LMIC setting.
Features repeated, age-adapted objective cognitive testing covering eight time points in early childhood alongside multiple waves in adulthood up to age .
Deployment of Latent Class Growth Analysis captures individual dynamic developmental velocities, avoiding reliance on arbitrary or transient single-point assessments.
Employs multiple imputation and mixed-effects modeling to account for clustered sibling data and missing item values.
Empirical findings align with and confirm trajectory patterns previously documented exclusively in high-income industrialized cohorts.
Study Limitations:
Attrition over half a century resulted in missing longitudinal data points, with only of original children meeting the minimum threshold of three childhood assessments.
Higher attrition rate among adult males due to domestic and international economic migration.
The extended recruitment interval () meant individuals entered the cohort at varying chronological ages, leading to missing infancy data for older birth years and missing school-age data for younger birth years.
The wide age span () necessitated grouping adulthood evaluations across broad chronological ranges.
LCGA remains an empirical methodology requiring investigator determination of class count.
Absence of unmeasured confounding variables, including direct parental psychometric intelligence scores, home stimulation quality, and caregiving interactions.
Implications for Research and Global Public Policy:
Timing of Targeted Interventions: The divergence of developmental trajectories prior to age , followed by rigid trajectory stability into school age, confirms the prenatal period and first as the definitive biological window for cognitive promotion.
Validity of Proxies: Using stunting and extreme poverty as sole proxies for developmental disadvantage fails to capture the full spectrum of cognitive trajectories; standardized, repeated cognitive tracking provides far superior diagnostic precision.
Mechanisms of Human Capital: Child cognitive development acts as a primary mediating conduit through which maternal nutritional status, physical capital (stature), and parental educational attainment translate into adult economic productivity, literacy, and abstract problem-solving capacity.
Breaking Intergenerational Poverty: Investments focused on maternal education, adolescent female nutrition, and prenatal care offer the highest leverage for halting the intergenerational transmission of socioeconomic disadvantage in transitioning economies.