1.1. The SES gradient across high- and low-income settings

Evidence quantifying the magnitude of these gradients has emerged primarily from high-income settings, where large longitudinal cohorts have made it possible to isolate and track their developmental consequences. In the United States, studies using large cohorts, like those funded by the National Institutes of Health (NIH), have shown that by the time they start school, children from lower SES backgrounds score lower on standardized cognitive tests than their peers from higher SES backgrounds [13]. These disparities are particularly notable in domains such as language and executive functions, with evidence suggesting that these gaps begin to emerge as early as 18 months of age [14]. Research also indicates that much of this gap growth occurs in early childhood, prior to school entry [15].

Likewise, cognitive and socioemotional development across the first ten years of life have been verified by studies from the UK Millennium Cohort Study and the Avon Longitudinal Study of Parents and Children (ALSPAC) [16,17]. Converging evidence from low- and middle-income countries (LMICs) has confirmed that these gradients are not confined to high-income settings alone. Multi-country analyses, including data from Bolivia and South Africa have documented that household SES predicts cognitive and language development from infancy through mid-childhood across highly diverse cultural and economic contexts, although with variation in magnitude across settings [18,19]. This reinforces that the link between lower developmental potential and family SES is a universal, cross-culturally consistent phenomena rather than a product of any particular environment.

1.2. Developmental timing and the home environment

Beyond establishing the universality of these gradients, literature has increasingly sought to identify the specific developmental windows during which SES-linked disparities first take root. Each childhood period presents different opportunities and challenges [20]. The early childhood years are vital for neurodevelopment, particularly for the formation of language and memory skills. Disparities in these skills between socioeconomic groups are well documented and generally emerge relatively early, typically by the second year of life [21]. Research underlines the importance of the home environment during this period; stable, rich home literacy environments (HLEs) and stimulating parent-child interactions significantly predict better future language and cognitive abilities [21–23].

Noble et al. established that HLE quality moderated the relationship between household socioeconomic status and newborn language development scores in the United States, indicating that the home environment is the primary mechanism by which socioeconomic disadvantage influences early neurodevelopment.[14]. Evidence from LMICs corroborate this conclusion: a prospective cohort study in rural Bangladesh revealed that caregiver stimulation and home learning resources accounted for a significant percentage of the SES gradients in language and cognitive scores. [24]. Similarly a study conducted in South Africa identified HLE quality as a key pathway through which household poverty translates into poorer early cognitive development [25]. However, maternal psychosocial factors such as depression and intimate partner violence have been shown to affect early childhood cognitive and behavioral development negatively [26–28].

1.3. Schooling, fadeout, and the persistence of disparities

Home-based influences, however, do not operate in isolation; as children approach school age, new environmental inputs begin to reshape their developmental trajectories.. High-quality preschool/kindergarten and primary school experiences can stimulate cognitive and academic achievement, potentially helping children catch up even if they displayed earlier delays [10,20,29,30]. In Jamaica, the Responsive Stimulation intervention, delivered to stunted infants aged 9–24 months produced sustained cognitive and psychosocial gains that persisted into adulthood [31]. However, entering a low-quality school or persistently experiencing low SES-related stress can also aggravate disparities [8,20]. In addition, early intervention gains in neurodevelopment may diminish without sustained support or quality follow-up, mainly for children with high cumulative risks [20,32]. This fadeout of early intervention effects is well documented in evaluations of the US Head Start program, where cognitive gains observed at kindergarten entry were substantially attenuated by Grade 3 [33]. Thus, while schooling offers recovery opportunities, structural inequities may hinder its compensatory effects, hence requiring multidisciplinary, sustained interventions targeting cognitive skills, executive functions, nutrition, and supportive home or school environments [34,35]. Whether the transition to formal schooling moderates SES-related neurodevelopmental trajectories in SSA contexts, and whether any such moderation is sufficient to alleviate the gradients established in early childhood, remains a critically under-examined question that this study addresses.

1.4. The sub-Saharan African and Tanzanian context

Although extensive research has explored the link between SES and neurodevelopment in high-income settings [1,8,11,36,37], a significant gap remains in understanding this relationship within low- and middle-income countries (LMICs). Sub-Saharan Africa (SSA) specifically bears the highest burden of children not reaching their developmental potential. This is largely due to unique contextual challenges like poverty, malnutrition, prevalent infections such as Human Immunodeficiency Virus (HIV) and malaria, violence, rapid socio-cultural shifts, and limited access to early childhood care and education [38–43], With each of these risk challenges carrying its own neurodevelopmental consequences. Tanzania specifically exemplifies a socioeconomic context that is marked by high disease burdens (e.g., HIV, malaria) and systemic inequities. Latest estimates suggest that 58% of children live below the poverty line, with restricted access to healthcare, education, a nutritious diet, and other essential resources [44–48].

While the negative impact of low SES background on neurodevelopmental outcomes seems well established, the magnitude of these differences seems to vary widely within SSA [41,45,49]. These differences are likely due to both differences in population and settings and differences in the neurodevelopmental instruments and outcome domain constructs used [34,41]. Data to compare children at different ages and at different developmental stages is scarce [34,41,49,50]. This measurement heterogeneity is a primary driver of the inconsistent and often incomparable findings across SSA studies and represents a major barrier to understanding how SES shapes neurodevelopment across distinct developmental stages.

Despite considerable success in adapting neurodevelopmental tools like the Bayley Scales of Infant and Toddler Development to specific countries and new tools directly developed in LMICs, such as the Malawi Development Assessment Tool (MDAT) and the Caregiver-Reported Early Development Instruments (CREDI), many local studies still rely on instruments that lack validation against the population’s linguistic and social realities [45,46,51,52]. The present study addresses this methodological gap by employing neurodevelopmental instruments that have undergone validation against the linguistic, cultural, and socioeconomic realities of the specific Tanzanian population under study [53].

1.5. Evidence gaps and the present study

Altogether, the literature reveals key gaps this study addresses. First, while SES–neurodevelopment associations in SSA are mostly cross-sectional, no Tanzanian study has examined how these evolve continuously from infancy to middle childhood, limiting inference on whether disparities widen, narrow, or persist. Second, although formal schooling is theorized to moderate SES-related disparities, this has not been tested in Tanzanian or East African community samples. Third, inconsistent SSA findings reflect measurement heterogeneity; this study uses locally validated neurodevelopmental tools, enabling valid cross-age comparisons and advancing regional evidence.This study explores the relationship between household wealth and childhood neurodevelopment in children aged 6 months to 10 years from Tanzania using instruments previously validated in this setting [53]. In doing so, we make two important contributions to the SSA neurodevelopmental literature: (1) characterizing, for the first time in this setting, how SES-related neurodevelopmental disparities manifest and evolve across three distinct developmental stages, i.e., early childhood, pre-school age, and school age; and (2) we test whether access to formal schooling moderates SES-related neurodevelopmental trajectories, providing evidence relevant to both developmental theory and educational policy in Tanzania and comparable LMIC contexts.

2.1. Study design and setting

This cross-sectional analysis leveraged baseline data from the Child Health and Infection with Low-Density Malaria (CHILD) randomized intervention study (ClinicalTrials.gov: NCT05567016; PACTR202311470737025). This parent trial investigates the long-term health and socioeconomic impacts of detecting and treating low-density malaria infections in children [54]. In the parent trial six hundred children aged 6 months to 10 years from Kiwangwa and Fukayosi wards in Tanzania’s Bagamoyo district were enrolled. The present analysis focused on baseline socioeconomic and neurodevelopmental data collected before trial interventions.

2.2. Participant selection and randomization

2.2.1. Participants selection.

We created household listings in the Kiwangwa and Fukayosi wards. From these listings, we randomly selected 660 children aged 6 months to 10 years to be invited for screening (Fig 1). To avoid data clustering and limit potential intervention contamination within families, we restricted eligibility to only one child per household. Initial screening and informed consent took place during home visits. Children were eligible if they were local residents, were not currently enrolled in other trials, and if their guardians agreed to seek care at the study clinic for illnesses and avoid outside medications, including herbal remedies [54].

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. A study flow chart.

Out of the 600 enrolled participants, seven were dropped due to incomplete neurodevelopmental assessments, and three were excluded due to data entry issues, leaving a final analysis set of 590 participants.

https://doi.org/10.1371/journal.pone.0354139.g001

We excluded children with a history of stage 3 or 4 HIV/AIDS, diabetes, or cancer, as well as those with a known allergy to artemether-lumefantrine (AL) or a weight below 5 kg [54]. Following the household screening, 600 eligible children were enrolled. Enrollment was finalized at the study health facility, where baseline assessments were conducted and participants were assigned to study arms. Participants were required to remain in the study for two years, though they could be withdrawn in cases of severe adverse events, loss to follow-up, or non-compliance with study protocols.

2.3. Consent procedures

Trained study staff obtained written informed consent from the parents or guardians of all potential participants before the commencement of any study procedures. The written consent forms, which included a statement about the possibility of future testing of collected samples, were provided in Swahili, the national language. Before the signing took place, staff verbally assessed the comprehension of the study details by the parents or guardians. In accordance with local guidelines, written minor assents were not obtained, as all study participants were under 13 years of age during the study period, which is the age threshold below which written assent is not standard practice.

2.4. Data collection procedures

All baseline assessments and interviews relevant to this analysis took place between 18 July and 29 November, 2023. Information was collected directly during neurodevelopmental assessments at the Kiwangwa or Fukayosi study clinics, and parent interviews at the clinics or during home visits. Doctors and nurses with significant training in delivering the neurodevelopmental batteries as outlined by Loss et al. conducted all examinations and data collecting [53]. The Open Data Kit (ODK) program was used to digitally gather data on Android-based tablets [56]. Castor Electronic Data Capture (EDC) was used to collect clinical data at the clinics located at the study sites. EDC is a web-based data capture system with real-time validation to ensure data quality and consistency [57]. To ensure continued adherence to research protocols,the study coordinator regularly supervised all study personnel.

2.5. Variables

We gathered demographic information on age, sex, village of residence, school information (ever attending school or not) and then grouped into childhood periods; early childhood (0.5–3 years), preschool age (4–5 years), and school age (6–10 years) as children are legally expected to begin Standard one of primary school at age 6 in Tanzania [58]. Socioeconomic information on presence or absence of the following household assets: electricity, ownership latrine/flush toilet, radio, TV, refrigerator, gas/electric stove, generator, bed(s), lamp(s), oven, hoe(s), sewing machine, motorbike, car, satellite dish, mobile phone, smartphone, and finished floors was also collected. Socioeconomic status (SES) was then assessed using a wealth index derived from Principal Component Analysis (PCA) of the household assets. Binary variables for each asset were included in the PCA to generate a composite factor score. Households were then categorized into tertiles (low, middle, and high SES) based on these scores to serve as a proxy for long-term household socioeconomic status.

Neurodevelopment was evaluated using age-appropriate instruments that assess core aspects of cognitive ability, which had been validated locally, specifically within the CHILD study context [53]. For children aged 6 months to <3.5 years, we used the standardized Global Scales for Early Development (GSED). The GSED was designed to objectively measure the motor, language, cognitive, and social-emotional development of children 0–3 years in low- and middle-income countries [59]. We administered the directly observed Long Form (GSED-LF), which has been validated in Tanzania. Assessments were conducted in Kiswahili, using the Tanzania-adapted version, at the Kiwangwa and Fukayosi dispensaries, by trained health professionals who had completed the standard GSED training protocol. Each child’s exact age was confirmed from their clinic card so that the tablet application could select the age-appropriate item set/ start point, and the caregiver was present throughout. Each assessment was conducted in a quiet, distraction-free space, with all materials prepared before testing began. In line with the standardized GSED procedure, assessors first established rapport with the child and caregiver, presented the tasks in a playful, low-pressure manner with consistent encouragement and breaks as needed, and asked caregivers not to prompt the child during testing. Responses were recorded directly on the GSED tablet application, which selects age-appropriate items and minimizes transcription errors, with assessments conducted under ongoing supervision to maintain consistency across assessors.

For children aged 3.5 to 5 years inclusive, we used the International Development and Early Learning Assessment (IDELA) battery, which directly assessed motor development, emergent literacy, emergent numeracy, and social-emotional development in preschool-aged children [60]. The following general intelligence and executive functioning constructs of the East Africa Cognitive Assessment Battery (EACAB) [61] were used to assess all school-aged study children (i.e., 6–10 years), except where otherwise stated: learning (6–8 years inclusive), planning, sequential processing, and two working memory constructs. In addition, school-aged children were tested for sustained attention using the pencil tap test (6–8 years inclusive) [62] and the code transmission sub-test of the Test of Everyday Attention for Children battery (TEA-Ch) [63].

All neurodevelopmental instruments were administered individually in a standardized, quiet setting by trained assessors. Administration times were approximately 10–25 minutes for GSED, 21–43 minutes for IDELA, and 45–60 minutes for the EACAB battery. To ensure data quality and minimize child fatigue, assessments were conducted in a single session with scheduled brief intervals between sub-tests.

GSED and IDELA instruments, by definition, provided a single overall functional score for each participant, which we standardized (z-scores) within each instrument for subsequent analysis. For the school aged children we generated composite scores using all domain tests available at the respective test age ranges (i.e., two test batteries for (i) 6–8 year olds: learning, planning, sequential processing, executive functioning, and sustained attention, and (ii) 9–10 year olds: planning, sequential processing, executive functioning, and sustained attention). We then used principal component analysis for the two age group batteries to derive the respective first principal component and generate standardized z-scores within each age battery. To derive the final global cognitive scores, results were pooled across all 50 iterations. The first principal component (PC1) explained an average of 49.17% of the variance in the 6–8years battery and 41.83% of the variance in the 9–10years battery. In both cases, the factor structure was highly stable across imputations, as evidenced by the negligible standard deviations in factor loadings. To adjust for the natural age-related progression of neurodevelopment, we predicted an age-residualized neurodevelopment score from a linear regression model entering age in years.

While tests were designed to minimize refusal, we found some refusals for sustained attention and executive function tests, especially at younger ages within the instrument group (refusals for 6–8 year olds: Pencil tap test: 7%, executive function working memory test 1: 8.21%, executive function working memory test 2: 19.%; refusals for 9–10 year olds: sustained attention test: 7.12%). Missing information on these sub-tests was filled in using the iterative Markov chain Monte Carlo multiple imputation method (using 50 imputed sets with Stata 18’s mi package). Specifically, we employed Multiple Imputation by Chained Equations (MICE) to generate 50 imputed datasets. Principal Component Analysis (PCA) was conducted separately within each of the 50 imputed datasets using a correlation matrix. The resulting first principal component (PC1) scores were then averaged across imputations to produce a single pooled estimate for each participant, which was subsequently z-score standardized for analysis.

2.6. Statistical analysis

We summarized characteristics of the study participants by counts and percentages or means and standard deviations as appropriate and used Student’s t-test or chi square test to test for differences between socioeconomic strata. SES of participants’ households was based on a wealth index, which was derived as the first principal component of available household assets [64–66]. Families were then grouped by their index into households with low (tertile 1), middle (tertile 2) or high (tertile 3) SES. To measure socioeconomic differences related to transitions into the formal schooling system, we calculated the proportions in school at ages 6−8 and kernel-weighted local polynomial regression analysis to visually explore non-linear relationships over age. We then explored the relationship of age‐residualized neurodevelopment scores and the extreme socioeconomic strata (high vs. low tertile) using local polynomial regressions and mean differences at each age-year group. Finally, to examine whether the association between SES groups and age-residualized neurodevelopment scores varied by childhood periods, early childhood (6 months −3 years), preschool age (4–5 years), and school age (6–10 years), we estimated a linear regression model adding a multiplicative interaction term for SES tertiles and childhood periods. We additionally added enumerator fixed effects to account for systematic variations in test administration.

Our primary objective was to document the magnitude of socioeconomic gradients. Thus, we included interviewers fixed effects in our regressions to control for potential observer differences but did not adjust for other factors (such as caregiver education) which might mediate the relationship between socioeconomic status and child development. We report main effects representing differences in residualized cognitive scores relative to individuals with high childhood SES. Analyses were performed using STATA 18 [67]. P-values of <0.05 were considered statistically significant, and 95% confidence intervals were reported. The complete analytical code is provided in S1 Text, and the de-identified minimal dataset is available in S1 Data.

2.7. Ethics statement

The study protocol has been reviewed and approved by Institutional Review Boards at University of California, San Francisco (Reference#: 342371), and Ifakara Health Institute, Tanzania (Reference#: IHI/IRB/No:08–2023), and by the Tanzania National Health Research Ethics Review Committee (Reference#: NIMR/HQ/R.8a/Vol.IX/4204). The study was approved by the President’s Office, Regional Administration and Local Government, Tanzania (PO-RALG) (Reference#: AB. 307/223/01). Additional information regarding ethical, cultural, and scientific considerations specific to inclusivity in global research is outlined in S4 File.

3.1. Household SES and school enrollment

At 6 years of age, we estimate 80% of children from the high SES families were in school, while only 45% of children from low SES families were enrolled in school (Fig 2). At 7 years of age, we estimate 90% of children from the High SES were enrolled in school, while only 50% of children from low SES families were enrolled in school at that age. Similarly, at 8 years of age, almost all the children from high SES families were enrolled compared to 80% of children from low SES families.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Local polynomial regression of children in school by age and wealth to explore potential enrollment effects.

The figure displays the estimated school enrollment rates for children from high- and low-socioeconomic status (SES) families across different ages.

https://doi.org/10.1371/journal.pone.0354139.g002

3.2. Household wealth and children’s neurodevelopment

Children from high SES households showed higher overall mean neurodevelopmental z-scores compared to their counterparts from low SES families (difference in age-residualized average z-scores across all ages: 0.25, P-value = 0.01, see details in S1 Table). Our non-linear model revealed an age-dependent emergence of neurodevelopmental gaps between socioeconomic status (SES) groups. While neurodevelopmental scores were similar at the youngest ages, they began to diverge around age 2 years in favor of high SES highlights (Fig 3: Panel A). Comparing mean differences per age-year group showed a similar result, where around age 4, the gap between the bottom and top tertile appeared to be stabilized at a level around 0.50 standard deviations, although statistically significantly only at age 8 years (Fig 3: Panel B).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Panel A. Smoothed plot of Neurodevelopment by Socioeconomic Status (SES). This figure is a smoothed plot that illustrates the absolute neurocognitive z-scores of children from high and low socioeconomic status (SES) groups across a range of ages. It shows a clear separation in scores, with the high SES group consistently scoring higher than the low SES group.

Panel B. Estimated Neurodevelopmental differences by Socioeconomic Status (SES). This figure displays the age-specific differences in neurodevelopmental scores between children from high- and low-socioeconomic status (SES) groups. The plot illustrates the estimated difference in scores at each year of age, with vertical lines representing the confidence intervals. It shows that the difference in scores becomes larger as children get older, indicating a widening gap between the two groups over time.

https://doi.org/10.1371/journal.pone.0354139.g003

At preschool age, children from low SES backgrounds statistically significantly scored on average 0.36 SD lower (95% CI: −0.64 to −0.08) than children from high SES backgrounds. In the school age group, similarly, the difference was 0.41 SD (95% CI: −0.73 to −0.09) (Table 2). In contrast, in the early childhood (ages 6 months −3 years) group, the estimated difference was small and not statistically significant. Differences between the high and middle SES groups were much smaller and not significant for any of the three developmental periods considered (Table 2, bottom panel). The full model and its derived strata-specific neurodevelopment score averages are detailed in the Appendix (see S2 Table and S3 Table).

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Socioeconomic status main effects on age-residualized neurodevelopmental scores (vs. the highest SES group) by schooling.

https://doi.org/10.1371/journal.pone.0354139.t002