English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Tools annual

Global: Zendy Free Plan

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Background The Early Childhood Education (ECE) Program, funded by the World Bank (USD 28 million), seeks to support expansion of quality ECE services with the objective of improving the development of children aged 3-5 years in disadvantaged communities in Lao PDR. The program targets an estimated 60,000 beneficiaries including children from 0-5 years, as well as their caregivers and wider communities throughout Northern Laos. Specifically, the program supports three different community-based interventions, the impact of which are being evaluated through three separate pragmatic clustered randomised control trials (RCTs) (Figre 1). Following completion of baseline data collection, program interventions commenced in November 2016, with endline due to be complete in April 2020. With an early education focus, improving children’s nutrition was not a key aim of the ECE program. However, midline results indicated a considerable reduction in prevalence of chronic undernutrition (i.e. stunting) amongst panel children between baseline and midline. Separate to the ECE program, school meal programs have been delivered in selected villages throughout Northern Laos, however, due to sample limitations (i.e. the number of schools with enough children and intervention budget) the condition of school meals was unable to be accommodated in the RCT design. For schools with a school meal program, school meals were expanded to include children aged 3-4 years attending CCDG and MAT interventions. Of methodological importance, the school meal program was a precursor to the ECE program and not all schools in the sample received the school meal program. Stunting, being too short for one’s age, is the result of chronic or recurrent undernutrition. Approximately 150 million children under 5 years (22%) experience stunting and associated cognitive impairments throughout the life course. Historically, the negative impacts of stunting have been thought to be irreversible after two years of age. However, there is dispute as to whether children who experience stunting before age two have the potential to “catch up” in height to their non-stunted peers, and if so, if those children also show cognitive catch up. Although provision of school meals was not included in the ECE study design, data collected provide opportunity to contribute to the debate around both physical and cognitive catch up growth. Evidence pertaining to this debate has the potential to inform early development and education strategies internationally. Traditionally, observational (i.e. non-randomised) data have been considered to exist within the domain of association and prediction, with causal inference preserved for studies with experimental (interventional or natural) design. More recently, advances in quantitative disciplines have brought development of new analytical techniques to draw causal conclusions from data collected in non-randomised settings (under a strict set of assumptions), supplementing traditional trial methods for causal estimation. These methods are grounded in the “potential outcomes” framework, and make use of “counterfactuals” and hypothetical trials to emulate experimental design within observational data. A counterfactual is a “what-if” statement drawing on the idea of causal contrasts within the same individual, e.g. if both exposure and non-exposure to a treatment could be observed for the same person, with all else held constant (i.e. both of their potential outcomes; Y1-Y0). Although it is not possible to observe both potential outcomes for an individual in the real world, this counterfactual approach employs sophisticated propensity weighting techniques and population level estimation to generate “exchangeable” groups of individuals, thus mimicking the randomisation and causal interpretability of a traditional trial. The present study seeks to utilise the aforementioned techniques in combination with comprehensive observational panel data collected through the ECE Program to determine if provision/receipt of school meals contributed to catch up in physical and cognitive growth of young children in Lao PDR. Methods Study sample The study sample included children at baseline and midline (n = 5680), who had an average age of 67 months at midline (5.6 years old), ranging from 37-90 months. Due to missing data in some of the key study variables (e.g. lack of known specific birth date), the final analysis sample included 4,948 children, representing 87% of the total sample of panel children. Exposure, outcome and confounders The key exposure was a binary variable measured via caretaker questionnaire, asking if their child had received meals from their ECE facility or primary school in the last week. Children’s height and weight was measured at baseline and midline, and caretakers provided children’s date of birth. Height-for-age z-scores (HAZ) were calculated using the WHO growth standards. The difference in HAZ scores from baseline to midline data collection was used as the key outcome of this study, with a binary variable indicating whether a child was stunted employed as a secondary outcome. Several variables were identified a priori as confounding the relationship between receipt of school meals and changes in HAZ scores/stunting, including child age, gender, and ethnicity, household socio-economic position, caretaker’s education, and village accessibility. Statistical analysis Descriptive statistics (prevalence) were computed for the outcome (stunted; yes, no) across each wave (baseline and midline), and stratified by child’s age at baseline. To determine the effect of receipt of school meals at baseline on stunting at midline, weighted linear and logistic regressions were employed, using stabilised inverse probability of treatment weighting (SIPTW) to control for observed confounding. The SIPTW can be represented in equation 1: Where A is the exposure (receipt of school meals) and C is a vector of observed confounders. The SIPTW approach makes use of propensity score weighting to generate a pseudo-population where the exposure is no longer related to the observed confounders, thus making the exposure groups “exchangeable” on these characteristics (mimicking randomisation). The final outcome models are represented in equations 2 and 3: Where E(Y) is the mean HAZ difference score, is the probability a child is stunted, is the intercept, is the coefficient for receipt of school meals, is the coefficient for the age difference between baseline and midline, is the coefficient for stunting at baseline and is an error term with Gaussian distribution in equation 2 and binominal distribution in equation 3. Results Table 1 shows that from baseline to midline there was a reduction in stunting of 8.43%. Table 2 indicates that the biggest reduction in stunting was observed for children aged 4 years old (12.0%), with smaller reductions observed for children aged 3 (6.4%) and 2 years old (6.2%). The receipt of school meals is hypothesised to be a potential driver of observed reductions in stunting from baseline to midline. Table 3 presents the distribution of children receiving school meals and reductions in stunting from baseline to midline, by child age. This indicates a possible relationship between the receipt of school meal and a reduction in stunting, with reductions in stunting larger in age groups with a higher proportion of children receiving school meals. This analysis is descriptive however, and does not take into account other characteristics that could also influence receipt of school meals and reductions in stunting, such as poverty. Regression analysis To isolate the relationship between receipt of school meals and reductions in stunting we employed linear and logistic regression models, weighted by SIPTW. This regression analysis indicates that receipt of a school meal at baseline results in a 0.07 standard deviation increase in height-for-age z-score at midline (Table 4). Results of the logistic regression indicate that children who received a school meal had 0.77 lower odds of being stunted at midline, after controlling for confounding, time between measurement at baseline and midline, and baseline stunting (Table 5). Methodological issue – effect interpretation In RCTs, the most common treatment effects calculated are the average treatment effect (ATE) and the average treatment effect on the treated (ATT). These are also known as intent to treat (ITT) and as-treated (AT) effects. The former indicates the effect of being assigned to a treatment on the outcome, whereas the latter indicates the effect of actually being treated on the outcome. As this study is attempting to replicate an RCT, a key question is, what is the most relevant effect estimate? Our results can be understood to be the ATT effect, as they represent the effect of receiving a school meal on stunting outcomes. In the RCT world, ATT/AT effects have been criticised due to potential external factors that affect choice of treatment, so the question is whether the present analysis has adequately accounted for these observed and unobserved confounding factors that affect self-selection into the treatment group i.e. receipt of school meals. Alternatively, the ATE/ITT effect is often considered closer to a real world effect of a treatment, however is strongly influenced by adherence. This can affect the generalisability of results, as different trials may have different adherence patterns. In relation to our trial, this effect could be estimated using aggregated village level data with the exposure re-coded to represent provision of school meals at ECE and school facilities within a village. However, school meals were employed in villages by multiple providers, thus defining school meal provision in a uniform way is difficult to ascertain. Finally, it may be appropriate to instead analyse the condition of school meals as a subgroup analysis within the original RCT design, despite the trial not having been powered to detect this effect. Conclusions The key conclusions of this abstract can summarised in terms of the central methodological questions that we are seeking to answer in collaboration with experts in the ECR workshop including: Key messages Data collected through the ECE Program in Laos provide opportunity to contribute to the debate around physical and cognitive “catch up growth” in stunted children. Using a counterfactual approach, our estimated as-treated effect indicated that stunted children aged 2-4 years at baseline who received school meals had lower odds of being stunted at midline (approximately two years later), however, we question if an intent-to-treat or sub-group analysis is better suited considering effects of interest and limitations in estimating effects.

795Contribution of school meals provision/receipt to “catch up growth” in young children: A counterfactual approach