Objective To determine the magnitude of relationships of early life factors with child development in low/middle-income countries (LMICs). Design Meta-analyses of standardised mean differences (SMDs) estimated from published and unpublished data. Data sources We searched Medline, bibliographies of key articles and reviews, and grey literature to identify studies from LMICs that collected data on early life exposures and child development. The most recent search was done on 4 November 2014. We then invited the first authors of the publications and investigators of unpublished studies to participate in the study. Eligibility criteria for selecting studies Studies that assessed at least one domain of child development in at least 100 children under 7 years of age and collected at least one early life factor of interest were included in the study. Analyses Linear regression models were used to assess SMDs in child development by parental and child factors within each study. We then produced pooled estimates across studies using random effects meta-analyses. Results We retrieved data from 21 studies including 20 882 children across 13 LMICs, to assess the associations of exposure to 14 major risk factors with child development. Children of mothers with secondary schooling had 0.14 SD (95% CI 0.05 to 0.25) higher cognitive scores compared with children whose mothers had primary education. Preterm birth was associated with 0.14 SD (-0.24 to-0.05) and 0.23 SD (-0.42 to-0.03) reductions in cognitive and motor scores, respectively. Maternal short stature, anaemia in infancy and lack of access to clean water and sanitation had significant negative associations with cognitive and motor development with effects ranging from-0.18 to-0.10 SDs. Conclusions Differential parental, environmental and nutritional factors contribute to disparities in child development across LMICs. Targeting these factors from prepregnancy through childhood may improve health and development of children.
We searched Medline, bibliographies of key articles and reviews, and grey literature to identify datasets from LMICs that collected data on early life exposures and child development. Search terms included a list of risk factors, terms related to motor, cognitive, language and socioemotional development, and a list of LMICs (list of search terms, online supplementary appendix 1). The most recent search was done on 4 November 2014. We also identified additional datasets via communication with researchers of published studies that were not retrieved in our search. The primary criterion for inclusion of the datasets was the assessment of at least one domain of child development (cognitive, motor, language and socioemotional) using a standard child development assessment instrument in at least 100 children before 7 years of age, as well as the collection of at least one early life factor of interest as part of the study. bmjopen-2018-026449supp001.pdf Following identification of the potential datasets, we contacted 50 first authors of the publications and investigators of unpublished studies, of whom 33 (66%) responded to participate in the present study (figure 1). We asked researchers to complete a survey that included questions about child development assessment tools used, age of developmental assessment and details on the early life factors measured in their study. Following the survey, 10 investigators declined to participate, 2 studies were excluded as the eligible sample size was <100 and 1 study was excluded as development was assessed after age 7 years. The investigators then shared results of predefined analyses on their data or shared data with researchers at the Harvard T H Chan School of Public Health to complete the analyses of individual studies and the meta-analyses. Flowchart of study selection. We created a list of early life risk factors based on the review of the current literature.13 14 These risk factors are represented in the ‘Good Health’ and ‘Adequate Nutrition’ components of nurturing care framework for ECD proposed by the WHO.17 We enquired about the availability of data on a list of risk factors in the preliminary survey sent to the investigators. Based on the survey responses, we then selected 14 early life factors that were available in at least four datasets to include in the pooled analyses. Following the standard definitions of categories used in published studies and the survey responses on how individual studies recorded data on each risk factors, we used uniform categorization of the risk factors applicable to all datasets. Risk factors were grouped into parental factors: father’s education and mother’s education (categories for each variable: none <1 year; primary 1 to <6 years; secondary 6 to <10 years; higher ≥10 years), maternal age (<15, 15 to <20, 20 to <35, ≥35 years), maternal height (<145, 145 to <150, 150 to 155 cm) maternal body mass index (BMI; <18.5, 18.5 to <25, 25 to <30, ≥30 kg/m2), haemoglobin level during pregnancy (normal ≥110 g/L; mild anaemia 100–109 g/L; moderate anaemia 70–99 g/L) and child factors: birth weight (low birth weight <2500 g; moderate low 2000–2500 g; very low birth weight <2000 g), preterm birth (preterm <37 weeks; late preterm 34–37 weeks; early preterm <34 weeks), small for gestational age (SGA; <10 percentile; moderate SGA 3 to <10 percentile; severe SGA <3 percentile) as determined by Alexander and Oken standards, exclusive breast feeding until 6 months of age, haemoglobin levels in infancy (normal ≥110 g/L; mild anaemia 100–109 g/L; moderate anaemia 70–99 g/L), access to clean water (yes, no), access to sanitation (yes, no) and diarrhoea preceding the 6 months before development assessment (yes, no). Details on the definition and categories of the risk factors are included in online supplementary appendix 2. We also enquired about data on birth spacing, maternal HIV infection, malaria, intimate partner violence and depression, but a limited number of studies had data on these factors. We included cognitive, motor and language outcomes in the analyses, socioemotional outcomes were not measured in a sufficient number of studies. If a study measured child development on multiple occasions, we included the measurement obtained at the age closest to 24 months. Since different tools were used for development assessment across studies, all development scores were standardised (z-scored) to ensure comparability between the measurements in different studies. Within each study, linear regression models were used to assess standardised mean differences (SMDs) in cognitive, motor and language scores for the selected risk factors. Multivariable models were adjusted for child’s age and sex, maternal education and a measure of socioeconomic status (eg, household income or wealth index). Maternal education was adjusted as a confounder in all models except for the model that estimated the effects of maternal education. If a study was a randomised trial, intervention assignment was also included in the adjusted model. In addition, estimates for preterm birth and gestation-specific birth weight category (SGA and appropriate-for-gestational-age) were adjusted for each other. The missing indicator method was used for covariates when 10% were missing the covariate was excluded from the analyses. Meta-analysis for a given risk factor was conducted if estimates from at least four studies were available. To account for the variation in tools used for measuring development, we only pooled the means and SEs of the standardised outcomes scores. As multivariable adjustment substantially changed the effect estimates, we used the adjusted effect estimates for meta-analysis. Given that heterogeneous effects seemed likely across the large variety of contexts studied, random effects meta-analysis was conducted using the DerSimonian and Larid method.18 Heterogeneity was assessed using I2 statistics. All analyses were conducted using the metaan commands in Stata V.12.0. Patients and or public were not involved.