Effects of responsive caregiving and learning opportunities during pre-school ages on the association of early adversities and adolescent human capital: an analysis of birth cohorts in two middle-income countries

Background: Millions of children globally are at risk of not reaching their developmental potential because of early adversities. We hypothesised that responsive caregiving and learning opportunities, components of nurturing care, at pre-school ages might mitigate the effects of adversities. Methods: We analysed longitudinal birth cohort data from Brazil (1993 Pelotas Birth Cohort, n=632) and South Africa (Birth to Twenty Plus [Bt20+] Birth Cohort, n=1130) to assess whether responsive caregiving and learning opportunities at pre-school ages (2–4 years) modified associations between cumulative early adversities and adolescent human capital. The cumulative adversities score (range 0–9) included household wealth and crowding; mothers’ schooling, height, age, and mental health; and children’s birthweight, gestational age, and length at age 12 months. We extracted data on responsive caregiving and learning opportunities from the Early Childhood Home Observation for Measurement of the Environment inventory, assessed at age 4 years (1993 Pelotas cohort) and 2 years (Bt20+ cohort). We examined three human capital indicators: intelligence quotient (IQ) assessed at age 18 years (1993 Pelotas cohort) and 16 years (Bt20+ cohort); psychosocial adjustment assessed at age 15 years and 14 years, respectively; and height assessed at age 18 years and 16 years, respectively. We used linear models with interaction terms between cumulative adversities, and responsive caregiving and learning opportunities, to predict adolescent human capital. Findings: For each additional Z score of total cumulative adversity, adolescent IQ decreased by 5·89 (95% CI −7·29 to −4·50) points in the 1993 Pelotas cohort (p<0·0001) and 2·69 (–4·52 to −0·86) points in the Bt20+ cohort (p=0·0039). After adjusting for total cumulative adversities, adolescent IQ points increased by 5·47 (95% CI 4·20 to 6·74) with each additional Z score of learning opportunities and by 2·26 (0·93 to 3·59) with each additional Z score of responsive caregiving in the 1993 Pelotas cohort, but not in the Bt20+ cohort (0·86 [–0·12 to 1·83] and 0·65 [–0·32 to 1·61], respectively). Associations between early adversities and IQ were modified by learning opportunities in the 1993 Pelotas cohort (beta coefficient for interaction 1·74, 95% CI 0·43 to 3·04; p=0·0092) and by responsive caregiving in the Bt20+ cohort (2·24, 0·94 to 3·54; p=0·0075). High nurturing environment attenuated the negative effects of early cumulative adversities on IQ. Interpretation: Early nurturing home environments protect young children against effects of early adversities on adolescent IQ, with long-term positive associations on adolescent cognition in two middle-income countries. Funding: Bill & Melinda Gates Foundation. We used longitudinal data from early childhood into adolescence from ongoing prospective birth-cohort studies in Brazil (1993 Pelotas Birth Cohort) and South Africa (Soweto, Johannesburg, Birth to Twenty Plus [Bt20+] Birth Cohort). The 1993 Pelotas Birth Cohort is an ongoing population-based study designed to evaluate health and development across the lifespan. Pelotas is in the south of Brazil, and has a population of approximately 350 000 inhabitants.11 Children from the 1993 Pelotas cohort grew up when the country was transitioning from primarily rural low income, to primarily urban middle income with a unified health-care system and, starting in 2001, a large conditional cash transfer programme.12 At the beginning of the study, the mortality rate was 21 per 1000 births.13 5249 of the 5265 babies born in 1993 in Pelotas were enrolled in the birth cohort.14 Some visits and measures were done in a random subsample for logistical and financial purposes.14 The cohort assessment done at age 18 years had a follow-up rate of 81·4%.15 Each assessment was approved by the Research Ethics Committee of the Federal University of Pelotas School of Medicine. Participants (and their mothers at the early ages) provided written informed consent at each stage of the study. Bt20+ is an ongoing longitudinal birth cohort study in the metropolitan area of Soweto, South Africa,16 with the overall aim to evaluate health and wellbeing of children growing up in a rapidly urbanising environment.17 Soweto is a township adjacent to the city of Johannesburg in the province of Gauteng, with an estimated population of more than 2·5 million people. The Bt20+ cohort represents the first generation of children born in a democratic system after the breakdown of the apartheid state. At the national level, there has been increased urbanisation; improved access to child and health-care services and technology; and reduced poverty after the expansion of several remedial social programmes.18 Inequality and poverty remain high, with 20% of the Soweto population currently living in households with income below the poverty line (based on the cost of a reference food basket, approximately US$47 per month).19 Enrolment into Bt20+ started during pregnancy (gestational age between 26 weeks and 40 weeks) among women living in Soweto in 1990. Eligible babies and their caregivers were enrolled in the study (3273 dyads) and have been followed up frequently. The assessment done when participants were aged 17 years had a follow-up rate of 70·6%.16 Ethical approval for the study was granted by the Committee for Research on Human Subjects at the University of Witwatersrand, South Africa, and consent was obtained from all participating women, with assent or consent also obtained from children and adolescents. We examined the modifying effects of nurturing care in the home on the association between cumulative adversities and adolescent human capital (panel). Outcome (human capital) 1993 Pelotas Birth Cohort Birth to Twenty Plus (Bt20+) Birth Cohort Effect modifier (nurturing care in the home) 1993 Pelotas Birth Cohort Bt20+ Birth Cohort Cumulative adversities 1993 Pelotas Birth Cohort Bt20+ Birth Cohort Human capital is defined as the education, training, skills, and health that contribute to economic and other forms of productivity and social integration.20 We included the domains of adolescent IQ, psychosocial adjustment, and height assessed with differing, but conceptually equivalent, methods. In the 1993 Pelotas cohort, IQ at age 18 years was assessed using four subtests of the Wechsler Adult Intelligence Scale (WAIS-III short form; n=4050), namely similarities, picture completion, arithmetic, and symbol coding. The WAIS-III score was adapted and standardised for Brazil and normalised for analysis. In the Bt20+ cohort, cognitive development was assessed with the Raven's Standard Progressive Matrices test at age 16 years (n=1373). This test measures non-verbal cognitive functioning and has been used widely, including in South Africa.21 We measured psychosocial adjustment with reverse scaling of internalising and externalising problems.22 In the 1993 Pelotas cohort, psychosocial problems were assessed at age 15 years with the Strengths and Difficulty Questionnaire (SDQ) parent version, a brief screening tool adapted and validated for Brazil.23 Mothers answered 20 questions representing internalising and externalising items about their adolescents' behaviours. Parental responses are valid for assessing adolescents' internalising and externalising behaviours.24 In the Bt20+ cohort, the 112-item Youth Self-Report25 was completed by adolescents at age 14 years. Behaviours were rated on a 3-point scale (from not true [score 1] to often true [score 3]). For this analysis, we used the 12-item anxiety (internalising) and 14-item aggression–oppositionality (externalising) items that are most comparable with the SDQ items. For both cohorts, the questionnaires referenced the preceding 6 months. We summed the reverse-scaled items to generate scores denoting positive psychosocial adjustment, consistent with the direction of the other two measures of human capital. Standing heights were assessed using stadiometers in the 1993 Pelotas cohort at age 18 years, and at age 16 years in the Bt20+ cohort, matching the ages of IQ measurement. Heights were converted to Z scores using age-specific and sex-specific WHO standards.26 In the 1993 Pelotas cohort, the 55-item Early Childhood Home Observation for Measurement of the Environment (EC-HOME) inventory was done in a subsample (n=632) at age 4 years.27 Items were scored as 0 or 1 if absent or present, respectively, as observed by trained data collectors or reported by mothers. Sample items were learning subscale27 (ie, “Child is helped to learn shapes and sizes at home”) and language subscale (ie, “The caregiver sings to the child daily”). We summed the two subscales in 18-item learning opportunities scores. The items evaluating responsiveness included verbal caregiver–child interactions (eg, “Parent converses with child at least twice during visit”). We summed the items to create 7-item responsive caregiving scores, as predetermined by the EC-HOME. In the Bt20+ cohort, responsive caregiving and learning opportunities were assessed at age 2 years (n=1838) with age-appropriate questions similar to EC-HOME. Responsiveness was based on a 6-item questionnaire from the EC-HOME completed by interviewer observation (eg, “Does the child appear happy, confident, and secure in the mother's presence?”). Learning opportunities were based on a 5-item questionnaire (eg, “Is there anything you are trying to teach your child at the moment?”). All measures were coded with higher scores denoting higher nurturance. Early adversities were mother–child assessments during the perinatal period, infancy, and early childhood in the 1993 Pelotas and Bt20+ cohorts, before assessments of the nurturing care components.2 Measures were chosen a priori based on developmental theory6 and availability in the two datasets. No available adversity measures were excluded. Each measure contributed equally to the index. The environmental cumulative adversities index included: low-income household (lowest two wealth quintiles within each site); low maternal schooling (grades of schooling attainment below 60th percentile); short maternal stature (<150·1 cm, representing −2 height-for-age Z score below international standards28); maternal age at child's birth (<18 years); poor maternal mental health (self-report questionnaire in the 1993 Pelotas cohort [>7 points] at child age 4 years,29 and 24-item Pitt Depression Inventory in the Bt20+ cohort [≥20 points30] at child age 6 months); and household crowding (more than three people per room, UN threshold). Each child’s environmental cumulative adversities index was summed, ranging from 0–6 points. The child cumulative adversities index included: low birthweight (<2500 g), preterm birth (<37 weeks), and stunted growth at 12 months (length-for-age Z score less than −2 relative to international standards26). The child cumulative adversities index ranged from 0–3 points. The total early cumulative adversities index is the sum of the environmental and child adversities (range 0–9), with higher scores denoting more adversities. We restricted analysis to participants with responsive caregiving and learning opportunities data (1993 Pelotas cohort n=632; Bt20+ cohort n=1838; in appendix 2 p 1). In the 1993 Pelotas cohort, the EC-HOME inventory was done in a random subsample14 that included all low birthweight children (n=510) plus 20% of remaining children. All analyses accounted for the sampling weights. In the Bt20+ cohort, missingness was due to high urban mobility, with no statistically significant differences in birth outcomes between children retained through adolescence versus lost to follow-up.31 Our final analytical datasets for IQ were 547 and 1081; for psychosocial adjustments were 632 and 767; and for height-for-age Z score were 539 and 1130 in the 1993 Pelotas and Bt20+ cohorts, respectively. Z scores were created to allow comparison across sites on variables with different metrics in regression models, including psychosocial adjustments, responsive caregiving, learning opportunities, and environmental, child, and total cumulative adversities scores. Constructs were comparable across datasets, but were not pooled due to measurement and contextual differences. Variables were standardised to not force linearity in models testing interaction between adversities scores and nurturing variables. in Appendix 2 (p 2) provides the standardised values in mean Z scores of the analytical sample in addition to the descriptive statistics for the 1993 Pelotas and Bt20+ cohort. Full information maximum likelihood estimation was used to account for missing information in the cumulative adversities score items (maternal height, mental health, and child length-for-age Z score at 12 months were missing for 415 [30%], 526 [38%], and 615 [44%] participants, respectively, in the Bt20+ cohort). Other cumulative adversities items had less than 2% missing data. Models using listwise deletion yielded similar results, indicating that data were missing completely at random (in appendix 2 pp 8–10). Continuous variables were tested for differences between included and excluded cases using linear regression models. In the 1993 Pelotas and Bt20+ cohorts, participants with valid pre-school data were very similar to those with missing data (in appendix 2 p 3). We observed differences between included and excluded cases in maternal age in both sites; therefore, we did a sensitivity analysis to account for such differences in maternal age. We used multivariable linear regression models with full information maximum likelihood to examine associations between early cumulative adversities and adolescent human capital controlling for child sex. Multivariable linear regression models were used to examine associations between responsive caregiving and learning opportunities and adolescent human capital, controlling for total cumulative adversities. To examine whether responsive caregiving and learning opportunities modified associations between cumulative adversities and adolescent human capital, we included interaction terms between the cumulative adversities scores (total, environmental, and child) and the nurturing home environment score. If the interaction term was statistically significant, we plotted the moderating variable (nurturing) as low, medium, or high (ie, −2 Z score, mean, +2 Z score, respectively), and tested the slope of the predictor variable (adversities) to identify the association driving the interaction. Our model specification checks, including assessment of model residuals, revealed that all normality assumptions were met with continuous outcomes. Effect modification by sex was tested in all models with a three-way interaction term among nurturing variables, cumulative adversities scores, and sex. These interaction terms were not statistically significant; thus pooled results are presented, adjusted for sex. Statistical analyses were done using Stata, version 15.1. The funders of the study had no role in study design, data collection, data analysis, data interpretation, or report writing. All authors had full access to all the data in the study and had final responsibility for the decision to submit for publication.