Scaling Early Childhood Development – What to read | September 2025

This month brings a wave of AI-focused publications, showing just how quickly the evidence base in early childhood development (ECD) services is expanding! Mehta et al. review advances in artificial intelligence (AI) and precision nutrition, and their potential to reshape maternal and child health in low and middle-income countries (LMICs). Traditional nutrition assessments — such as anthropometry, biochemical markers and clinical exams — are labour-intensive, while AI offers faster, more accurate and scalable approaches. By analysing large datasets, AI can predict individual responses and tailor intervention. Although results from high-income countries are promising, generalisability to LMICs is uncertain due to differences in population characteristics such as metabolomes, microbiomes and genetics. Barriers to implementing nutrition precision modelling approaches in LMICs include poor data quality, weak infrastructure, low digital literacy and privacy concerns. Emerging AI tools – such as body scanners and dietary apps – demonstrate potential for scalable solutions, but further research on contextual adaptation and integration into routine programmes is essential to achieve successful scale-up and improve maternal and child health outcomes.

El Arab et al. reach a similar conclusion, synthesising evidence from 39 reviews on the integration of AI in maternal, neonatal, reproductive and ECD care. Applications span pre-conception (e.g. IVF workflows), pregnancy (e.g. foetal imaging), postpartum wellness (e.g. depression screening) and child development. While models often demonstrate high performance in research settings, most tools lack external validation and are rarely deployed in low-resource community settings. Examples such as smartphone image analysis that enabled community health workers in Indian villages to screen for anaemia without labs and wearable sensor networks with on-device analytics that monitored blood pressure and heart rate to improve preeclampsia care in rural Ghana, illustrate how AI can be made affordable and scalable in LMICs. The findings highlight the potential for AI to strengthen maternal and child health care – but only if implementation is accompanied by investment in digital infrastructure, ethical oversight and health-worker readiness.El Arab et al. reach a similar conclusion, synthesising evidence from 39 reviews on the integration of AI in maternal, neonatal, reproductive and ECD care. Applications span pre-conception (e.g. IVF workflows), pregnancy (e.g. foetal imaging), postpartum wellness (e.g. depression screening) and child development. While models often demonstrate high performance in research settings, most tools lack external validation and are rarely deployed in low-resource community settings. Examples such as smartphone image analysis that enabled community health workers in Indian villages to screen for anaemia without labs and wearable sensor networks with on-device analytics that monitored blood pressure and heart rate to improve preeclampsia care in rural Ghana, illustrate how AI can be made affordable and scalable in LMICs. The findings highlight the potential for AI to strengthen maternal and child health care – but only if implementation is accompanied by investment in digital infrastructure, ethical oversight and health-worker readiness.

Benson et al.’s new scoping review maps existing literature on AI in ECD research more broadly, identifying key gaps and priorities for future research. Interest in machine learning is growing, particularly for predicting developmental outcomes, but the evidence base is limited. Of 27 studies, 78% were from high-income countries and none from sub-Saharan Africa, underscoring a major geographical imbalance. Most studies focused on risk prediction after age 2, missing opportunities for earlier intervention. Across studies, model validation was inconsistent, external validation rare, sample sizes small and datasets imbalanced. Only one study reported practical use, highlighting the gap between tool development and real-world implementation. The authors call for future research to collect comprehensive, longitudinal and contextually relevant data – especially in LMICs – to ensure ML tools are both practical and culturally relevant.

Turning to parenting, a new discussion paper by Ogutu highlights barriers and opportunities for scaling gender-transformative programmes to address family violence in Kenya. While many NGO-led pilots have shown promise, few achieve government adaptation or scale. As one official noted, ‘I’ve yet to see an NGO program that the government truly embraces, especially a parenting program. Many […] come from the West, and […] don’t align with our African traditions.’ Ogutu traces the successful Kenya Positive Parenting Programme´s path from NGO initiative to nationally budgeted intervention, crediting sustained government engagement, use of domestic violence data and field visits with government officials to showcase impact. Letting go of ownership, building broad coalitions and creating scaling roadmaps with multisectoral teams were also critical. The study offers rare insights into the real-world challenges of achieving genuine, sustained scale-up of ECD initiatives.

In a similar effort to provide practical guidance on scaling ECD, Wolf et al. contribute a chapter to Global Education: Linking Theory and Practice, which details the successes and challenges of scaling a successful teacher in-service training programme in Ghana. The study stresses the importance of a partnership between the research team and Ghana’s Ministry of Education. Working with the National Nursery Teacher Training Centre and Sabre Education, the team adapted an existing teacher training programme for simplicity and scalability. An impact evaluation confirmed its effectiveness, showing reduced teacher burnout and dropout. Yet, despite extensive policy engagement and integration into national standards, the programme´s implementation remains limited.

Malik et al. conducted a scoping review on community-led monitoring approaches — community involvement in monitoring public services — to inform a monitoring, evaluation, accountability, and learning (MEAL) framework for HIV programmes. The review found that while community engagement in monitoring is increasing, systematic, routine evaluation of community-led monitoring effectiveness remains limited. Most initiatives rely on one-off assessments, with sustainability challenged by fragmented donor funding and unpaid volunteers. Participatory, people-centred monitoring shows greater potential for scale-up and lasting impact. The authors argue that community-led monitoring can enhance accountability, service quality and community engagement – but these outcomes are hard to capture with traditional MEAL models. They recommend adaptive, participatory MEAL frameworks that value community narratives, positioning community-led monitoring as an essential complement to government-led monitoring.

In the last half of this newsletter, we spotlight four new studies that uniquely leverage large longitudinal datasets — still rare in ECD research — to track trends in ECD over time. Elisaria et al. from the Thrive team analyse 30 years of demographic and health survey data to track and describe child stunting trends in Tanzania, showing a decline from 50% in 1992 to 30% in 2022. However, stunting rates remain higher than the global average and vary significantly by region and socioeconomic status. Risk factors include low birthweight, maternal undernutrition and limited education, short birth intervals, poor sanitation and poverty; protective factors include maternal schooling and higher household wealth. The study highlights that further progress will require multisectoral strategies that tackle structural inequalities alongside maternal and child health interventions. The findings underline a multisectoral approach to ECD programmes, particularly to create impact at scale.

Basterra et al. present the largest longitudinal study on the combined effects of all types of social protection programmes on child health, covering 46 LMICs and a third of the global under 5 population (2000–2021). Social protection coverage – including cash transfers, subsidies and in-kind support – was strongly positively associated with child health: for every 1% increase in social protection coverage, there was a 0.34% reduction in under 5 mortality, a 0.29% reduction in stunting and a 0.14% reduction in wasting. The authors estimate over 3 million under 5 deaths were averted, including 583,590 during the pandemic, with especially strong effects for toddlers and girls. They conclude that scaling up comprehensive social protection should be prioritised as a key strategy for child survival, resilience and equity.

The Young Lives project — internationally recognised for being one of the largest and longest-running longitudinal studies on childhood poverty and development in LMICs (tracking over 12,000 children across Ethiopia, India, Peru, and Vietnam) — has released a new policy brief on youth mental health. Highlighting crises such as COVID-19, the climate emergency and conflict in Ethiopia, the research documents rising rates of depression, anxiety, PTSD and psychosis. The team calls for cross-cutting solutions, including stronger social protection, early childhood and parental support, and measures to tackle gender inequalities like child marriage and unpaid care. They argue for coordinated, multisectoral action across health, welfare and education to build more sustainable and scaled responses.

Finally, Katus and Fox review 25 years of longitudinal research on infant neurodevelopment, stressing the need to expand studies and interventions. Landmark longitudinal studies using neuroimaging in Jamaica and Romania showed that early interventions – nutritional support, psychosocial stimulation and foster care – can yield lasting cognitive, social and neural benefits. Recent projects in The Gambia, India, Bangladesh, Brazil and South Africa use portable neuroimaging (electroencephalography (EEG), functional near-infrared spectroscopy (fNIRS)) to identify early biomarkers of brain development. The review highlights promising approaches (such as maternal support, family-based training, financial transfers and nutritional supplementation) and calls for culturally appropriate, sustainable programmes that embed neural markers into health care and policy to help all children thrive in their first 1,000 days.