Evidence is needed on effective approaches to build parents’ ability to promote child development feasible in low- and middle-income countries. Our objective was to synthesize impact of the Reach Up early childhood parenting program in several low- and middle-income countries and examine moderation by family and implementation characteristics.
Systematic search using PubMed and Academic Search Elite/EBSCO Host. Randomized controlled trials of the Reach Up program from 1985 to February 2022 were selected. Data were extracted by 2 independent researchers. Primary outcomes were child cognitive, language, and motor development. Secondary outcomes were home stimulation and maternal depressive symptoms. We synthesized pooled effect sizes using random effect inverse-variance weighting and effect modification by testing pooled subgroup effect estimates using the χ2 test for heterogeneity.
Average effect size across 18 studies ranged from 0.49 (95% confidence interval [CI] 0.32 to 0.66) for cognition, 0.38 (CI 0.24 to 0.51) for language, 0.27 (CI 0.13 to 0.40) for motor development, 0.37 (CI 0.21 to 0.54) for home stimulation, and –0.09 (CI –0.19 to 0.01) for maternal depressive symptoms. Impacts were larger in studies targeted to undernourished children, with mean enrollment older than age 12 months and intervention duration 6 to 12 months. Quality of evidence assessed with the Cochrane Assessment of Risk of Bias and GRADE system was moderate. Instruments used to assess child development varied. In moderator analyses, some subgroups included few studies.
Reach Up benefits child development and home stimulation and is adaptable across cultures and delivery methods. Child and implementation characteristics modified the effects, with implications for scaling.
Poor development in children younger than age 5 years in low- and middle-income countries (LMIC) is an enormous public health problem, with life-long consequences for individual educational attainment and future income and implications for national development.1 Poverty is a major driver of poor child development mediated, in part, by home environments with limited caregiver–child interaction and stimulation.1,2 Parenting interventions aimed at building caregiver skills to provide responsive interactions and stimulating home environments have benefited children’s development and parenting behaviors,3–6 and there is some evidence of sustained benefits.7 The challenge remains taking programs to scale with a need for evidence on family and program implementation characteristics that may modify effectiveness.
In this review, we examine the impact of interventions based on the Jamaica stimulation intervention designed in the 1970s–1980s and aimed to build parents’ skills to promote child development.8,9 The Reach Up early childhood parenting program is based on the curriculum for the Jamaica stimulation intervention; additional manuals and materials were developed to support program implementation, including adaptation, training, and supervision.9,10 Implemented in several LMIC, interventions range from small efficacy trials to large-scale implementation, including national programs. The model is also the only 1 developed in LMIC with evidence of long-term gains.11 By focusing on 1 model, we are able to examine how impact is influenced by family and implementation characteristics, which is critical to guide the design and implementation of programs at scale. Previous reviews have been limited in their ability to do this because of wide variability in content and design of interventions examined.6
The objectives are to examine (1) the impact of the Jamaica stimulation intervention/Reach Up early childhood parenting program on early childhood development (ECD), home stimulation, and maternal well-being; (2) whether intervention affects child development or parent outcomes are moderated by characteristics of child and caregiver; and (3) how intervention impact is affected by implementation characteristics.
Methods
We examined randomized controlled trials (RCTs) based on the principles, content, and methods of the Jamaica stimulation intervention/Reach Up, a play-based intervention that builds parents’ ability to promote child development.8 The program uses a structured curriculum designed to facilitate delivery by persons with a minimum of complete primary education. Emphasis is placed on building the delivery staff’s relationship with parent and child, parent–child interaction and responsiveness, use of language, and praise for efforts and achievements (Supplemental Table 6). The Reach Up program provides weekly and fortnightly curricula for children aged 0 to 48 months, detailed training and supervision manuals, and guidance on adapting for context.10 Originally delivered through home visits, the intervention has also been adapted for delivery through small groups.12–14
We follow the Preferred Reporting Items for Systematic review and Meta-Analysis Protocols (PRISMA) 2020 statement and extended checklist. The protocol was registered with the International Prospective Register of Systematic Reviews, registration number CRD42020206313. Ammendments to the protocol are shown in Supplemental Table 9.
Eligibility Criteria
Original trials, published in English or Spanish, were included if the intervention was delivered to parents and children with enrollment aged 0 to 36 months. Studies with additional intervention arms (eg, nutrition) were included only if the effect of the stimulation component could be evaluated separately. Additionally, trials were only included if they included at least 1 of the primary outcomes of child cognitive, language and motor development. Secondary outcomes were home stimulation and maternal depressive symptoms. We considered child behavior and other aspects of parent well-being, but sufficient data were not available. Peer-reviewed articles and working papers/preprints available in the public domain were considered.
Search Strategy and Study Selection
We developed an electronic search strategy in collaboration with a liaison health sciences librarian with expertise in knowledge synthesis. Two researchers (J.C.H. and J.S.) conducted independent searches covering the period January 1985 to October 2020. The search strategy was piloted using PubMed, Science Direct, and Academic Search Elite/EBSCO Host. A few modifications were made to search terms and, because of the limited number of Boolean/Phrase connectors possible in ScienceDirect, this database was excluded. A search was also done of electronic databases of The World Bank and Inter-American Development Bank. The search terms are listed in Supplemental Table 8. The search was repeated in February 2022 to identify studies published between October 2020 and February 2022.
Titles and abstracts of articles identified were screened independently by both researchers and discrepancies resolved through consensus. Full texts of articles selected were read to confirm eligibility for study inclusion. Discrepancies were resolved through consensus and 2 researchers in consultation with an arbitrator (S.W.).
Data Extraction
Data were extracted independently by 2 researchers (J.C.H. and H.O.P.) using structured forms designed for this review. Discrepancies were resolved by consensus and an arbitrator (S.W.), when necessary. Data extracted included the study title, publication year, country, sample size in the intervention and control groups, timelines for measurements (baseline and endline), instruments used to measure outcomes, means and SDs, effect size, P values and adjustment covariates for the outcome measures of child development, home stimulation, and maternal depressive symptoms. We also extracted information on sample characteristics such as maternal education, child age on enrollment and nutritional status, and implementation information such as frequency, duration, training, supervision, and characteristics of delivery staff (defined as home visitors/group facilitators). Information required for analysis but not included in the article was requested from authors.
Assessment of Study Quality
The quality of the included studies was assessed independently by 2 researchers (J.C.H. and H.O.P.) using the Cochrane risk of bias tools for randomized and cluster-randomized trials to assess risk of bias from the randomization process, identification of recruitment participants, deviations from the intended intervention, missing outcome data, measurement of the outcome, and selection of the reported results. The risk of bias judgment for each randomized controlled trial was scored as “low,” “high,” or “some concerns” (Supplemental Table 9). The quality of the evidence was then evaluated using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) system, on the following components: risk of bias, consistency of results across studies, indirectness, imprecision, and publication bias.15
Statistical Analysis
For each study, effect estimates of endline mean differences comparing intervention to control were transformed to standardized mean differences (SMD; ie, effect sizes) using Hedges’ g with a pooled SD if available16 ; otherwise, they were standardized with control group or baseline SDs. For cluster-randomized trials, sufficient information about the clustering and intracluster correlation is often unavailable. Consequently, we calculated a cluster adjusted Hedges’ g only when there was a substantive disparity between the study reported effect size and Hedges’ g. Pooled effect sizes were then synthesized using a random effect inverse-variance weighting approach and heterogeneity was assessed by I2 statistics. Because studies use instruments that either report cognition alone (eg, Bayley Scales of Infant and Toddler Development [Bayley]-III) or combined cognition and language (eg, Bayley-II Mental Development Index, Regional Project on Child Development Indicators), we present analyses for (1) studies that report cognition alone and (2) studies that report cognition alone plus studies that report combined cognition and language. We use this approach also for language and motor development. In the former, some studies report language alone and some report combined cognition and language. In the latter, some studies report fine and gross motor development alone and some report them combined.
Analysis of Subgroups
Subgroup analyses included characteristics that the literature suggested may affect intervention impact and data were available. Individual-level heterogeneity analysis by maternal education was able to be assessed within each study. We use 2 cut points to define maternal education groups. The first splits into groups of mothers/caregivers who have primary level or less education or mothers/caregivers with more than primary school education. The second splits the groups into mothers/caregivers in the lower half of education for each study sample or in the upper half of education. Child characteristics evaluated at study level included mean child age on enrollment and whether the intervention was targeted to undernourished children. Study implementation characteristics evaluated include duration of intervention, frequency of contacts, size of program, delivery by home visits or mother/caregiver and child groups, whether the intervention is a standalone program or home visitors/group facilitators are paid by government, frequency of supervision provided to staff delivering the intervention (observations of home visits or group sessions), and urban or rural location. Effect modification was assessed by using the intervention effect estimates from each of the effect modifier subgroups. These estimates were pooled within subgroup to generate the pooled subgroup effect estimate and heterogeneity between subgroups was assessed using the χ2 test for heterogeneity.
Results
Figure 1 presents the flow diagram of the identification, screening, and selection of RCTs. A total of 2029 records were identified from the search of the databases. Three additional articles were identified based on communication from review authors. We excluded 1986 records based on titles and abstracts. The full text of 46 papers was reviewed and 28 excluded because they did not meet eligibility criteria, leaving a final sample of 18 studies.
PRISMA flow diagram of the identification, screening, and selection of RCTs in this review.
PRISMA flow diagram of the identification, screening, and selection of RCTs in this review.
Studies are summarized in Table 1 and information on primary and secondary outcome measures presented in Table 2. Thirteen studies were cluster randomized and 5 individually randomized. Three studies had 2 parallel intervention arms,13,31,38 giving a maximum of 21 trials.
Summary of the Studies in Alphabetical Order by Author
| Author, Country | Study Design/No. of Clusters | Trial Description | Delivered Through Government System | Frequency and Modality | Enrolled Sample | Analysis Samplea | Age at Enrollment (range, mo) | Program Duration (mo) |
|---|---|---|---|---|---|---|---|---|
| Andrew et al,25 India | Cluster RCT 27 slums per trial arm | • Periurban slums in Cuttack • Intervention delivered by local women hired for the project | No | Weekly home visits | Intervention = 209 Control = 212 | Intervention = 191 Control = 187 | 10–20 | 18 |
| Araujo et al,21 Peru | Cluster RCT 60 districts per trial arm | • Rural districts with high levels of poverty and stunting • Districts assigned to first wave received intervention and districts assigned to second wave were the controls • Intervention delivered by local women hired for the Cuna Mas program | Yes; Cuna Mas | Weekly home visits | Intervention = 3894 Control = 2003 | Intervention = 3192 Control = 1493 | 0–24 | 24 |
| Attanasio et al,23 Colombia | Cluster RCT 24 areas per trial arm | • Rural municipalities in Bogotá • 4-arm trial: psychosocial stimulation (PS), micronutrient supplementation (MS), psychosocial + supplementation (PS + MS), and control. • Intervention delivered by mother leaders on cash-transfer program hired to the project | No | Weekly home visits | Intervention (PS) = 360 Control = 351 | Intervention (PS) = 318 Control = 318 | 12–24 | 18 |
| Brentani et al,31 Brazil | Individual RCT Parallel trial | • Urban areas of Sao Paulo • Parallel trials: community health workers (CHW) delivered Intervention in areas covered by the national primary care program; new cadre of child development agents (CDA) delivered intervention in areas not covered by the program • Excluded children already enrolled in full-time day care at baseline | Yes; national home-based primary care program | Fortnightly home visits | Intervention (CHW) = 164 Control (CHW) = 164 Intervention (CDA) = 249 Control (CDA) = 249 | Intervention (CHW) = 145 Control (CHW) = 149 Intervention (CDA) = 211 Control (CDA) = 215 | 9–17 | 12 |
| Galasso et al,34 Madagascar | Cluster RCT 25 regions per trial arm | • Rural regions • 5-arm trial: control (T0), intensive nutrition counseling (T1), T1+ lipid-based nutrient supplementation (LNS) to children aged 6–18 mo (T2), T1+ LNS to pregnant women, lactating mothers of children aged 0–5 mo and children aged 6–30 mo (T3) and T1+ stimulation to children aged 6–30 mo (T4) • Intervention delivered by community health workers hired to the project | Yes; national nutrition program | Fortnightly home visits | Intervention (T4) = 750 Control (T0) = 747 | Intervention (T4) = 719 Control (T0) = 732 | 0–11 | 24 |
| Grantham-McGrego et al,8 Jamaica | Individual RCT | • Poor urban neighborhoods in Kingston • 4-arm trial: supplementation only, stimulation only, supplementation + stimulation, and control • Targeted stunted children (height-for-age <–2 SD) • Intervention delivered by community health workers hired for the project | No | Weekly home visits | Intervention (stim only) = 30 Control = 33 | Intervention (stim only) = 30 Control = 33 | 12–24 | 24 |
| Grantham-McGregor et al,13 India | Cluster RCT 48 villages per trial arm | • Rural villages in Cuttack, Salepur and Bolangir • 4-arm trial: control, nutritional education, nutritional education + home visits, nutritional education + group sessions • Intervention delivered by local women hired to the project | No | Weekly home visits or weekly group sessions of 7–8 mother–child pairs | HVs + nutr. edu. = 357 Groups + nutr. edu. = 346 Control = 353 | HVs + nutr. edu. = 332 Groups + nutr. edu. = 323 Control = 320 | 7–16 | 24 |
| Hamadani et al,26 Bangladesh | Cluster RCT 10 community nutrition centers per trial arm | • Poor rural area of Monohardi subdistrict • Targeted children with moderate and severe undernutrition (wt for age <–2 SD) • Intervention delivered by local women hired to the project | No | Group meetings: weekly for 10 mths and fortnightly for two mths Home visits: twice weekly for 8 mths and weekly for 4 mths | Intervention = 104 Control = 102 | Intervention = 92 Control =101 | 6–24 | 12 |
| Hamadani et al,12 Bangladesh | Cluster RCT 45 community clinics per trial arm | • Rural sub-districts in Narsingdi • Targeted underweight children (wt for age < -2 SD) • Intervention delivered by government health workers at community clinics | Yes | Group sessions (pairs of mothers) | Intervention = 358 Control = 360 | Intervention = 343 Control = 344 | 5–24 | 12 |
| Heckman et al,29 China | Cluster RCT 55 villages per trial arm | • Rural villages in Hauchi County • Targets disadvantaged families • Intervention delivered by home visitors hired by the project | No | Weekly home visits | Intervention =715 Control = 852 | Intervention = 541 Control = 547 | 0–24 | 22 |
| Hossain et al,35 Bangladesh | Cluster RCT 11 areas per trial arm | • Poor rural sub-district of Ullapara • 3-arm trial: psychosocial stimulation (PS) + unconditional cash transfer (UCT), UCT only and standard care group (comparison). • Intervention delivered by female village health workers hired by project | Yes; Government maternity allowance program | Fortnightly home visits | PS+UCT = 197 UCT only = 188 | PS+UCT = 182 UCT only = 179 | 6–16 | 12 |
| Mehrin et al,14 Bangladesh | Cluster RCT 20 community clinics per trial arm | • Rural district of Kishorganji • Children with weight for age < –1.5 SD, not hospitalized or requiring constant monitoring, and lived within 30-min walk from clinic • Intervention delivered by government health workers at community clinics | Yes | Fortnightly group sessions Groups of 4 mother– child pairs | Intervention = 419 Control = 366 | Intervention = 396 Control = 319 | 5–23 | 12 |
| Nahar et al,27 Bangladesh | Individual RCT | • Urban slums in Dhaka city • 4-arm trial: psychosocial (PS), food supplementation (FS), psychosocial + food stimulation (PS + FS), clinic control (CC) and hospital control (CH) • Targeted severely malnourished children (weight for age <–3 SD) without acute infections or requiring hospitalization • Intervention delivered by local women hired for the project | No | Fortnightly One or more mother–child pair at community clinic | Intervention (PS) = 102 Control (CC) = 99 | Intervention (PS) = 59 Control (CC) = 59 | 6–24 | 6 |
| Powell et al,36 Jamaica | Individual RCT | • Poor urban area in Kingston • Intervention delivered by community health workers from neighborhood clinic • 2 studies are reported; however, only study 2 (an RCT) is included in this evaluation | Yes | Weekly home visits | Intervention = 29 Control = 29 | Intervention = 29 Control = 29 | 16–30 | 12 |
| Powell et al,37 Jamaica | Cluster RCT 11 (intervention) and 7 (control) nutrition clinics | • Urban areas of Kingston and St Andrew • Targeted undernourished children weight for age <–1.5 SD and <–2 SD in the past 3 mo • Intervention delivered by community health aides from primary health nutrition clinics | Yes | Weekly home visits | Intervention = 70 Control = 69 | Intervention = 65 Control = 64 | 9–30 | 12 |
| Tofail et al,38 Bangladesh | Cluster RCT 15 villages per trial arm | • Rural villages in Monohordi subdistrict. • Parallel trials: children who had iron deficiency anemia (IDA) and children neither anemic nor iron deficient (NANI) • In addition to the psychosocial stimulation, iron syrup was given to children with IDA for the first 6 mo • Intervention delivered by local women hired to the project | No | Weekly home visits | Intervention (IDA) = 117 Control (IDA) = 108 Intervention (NANI) = 106 Control (NANI) = 103 | Intervention (IDA) = 110 Control (IDA) = 106 Intervention (NANI) = 104 Control (NANI) = 92 | 6–24 | 9 |
| Walker et al,17 Jamaica | Individual RCT | • Urban areas of Kingston and St Andrew • Targeted term low-birth-weight infants, gestational age ≥37 wk, birth weight <2500 g • Intervention: phase 1: first 8 weeks after birth; phase 2: from age 7 mo to 24 mo • Intervention delivered by community health workers hired for the project | No | Weekly home visits | Intervention = 70 Control = 70 | Intervention = 63 Control = 67 | Birth | 19 |
| Walker et al,39 Jamaica | Cluster RCT 5 centers per trial arm | • Children attending health centers in urban areas of Kingston and St Andrew • 4 trial arms: health center, home visits, health center + home visits, and control • Delivered by health center community health workers | Yes | Fortnightly home visits | Intervention (home visits) = 50 Control = 150 | Intervention (home visits) = 38 Control = 123 | 6 | 12 |
| Author, Country | Study Design/No. of Clusters | Trial Description | Delivered Through Government System | Frequency and Modality | Enrolled Sample | Analysis Samplea | Age at Enrollment (range, mo) | Program Duration (mo) |
|---|---|---|---|---|---|---|---|---|
| Andrew et al,25 India | Cluster RCT 27 slums per trial arm | • Periurban slums in Cuttack • Intervention delivered by local women hired for the project | No | Weekly home visits | Intervention = 209 Control = 212 | Intervention = 191 Control = 187 | 10–20 | 18 |
| Araujo et al,21 Peru | Cluster RCT 60 districts per trial arm | • Rural districts with high levels of poverty and stunting • Districts assigned to first wave received intervention and districts assigned to second wave were the controls • Intervention delivered by local women hired for the Cuna Mas program | Yes; Cuna Mas | Weekly home visits | Intervention = 3894 Control = 2003 | Intervention = 3192 Control = 1493 | 0–24 | 24 |
| Attanasio et al,23 Colombia | Cluster RCT 24 areas per trial arm | • Rural municipalities in Bogotá • 4-arm trial: psychosocial stimulation (PS), micronutrient supplementation (MS), psychosocial + supplementation (PS + MS), and control. • Intervention delivered by mother leaders on cash-transfer program hired to the project | No | Weekly home visits | Intervention (PS) = 360 Control = 351 | Intervention (PS) = 318 Control = 318 | 12–24 | 18 |
| Brentani et al,31 Brazil | Individual RCT Parallel trial | • Urban areas of Sao Paulo • Parallel trials: community health workers (CHW) delivered Intervention in areas covered by the national primary care program; new cadre of child development agents (CDA) delivered intervention in areas not covered by the program • Excluded children already enrolled in full-time day care at baseline | Yes; national home-based primary care program | Fortnightly home visits | Intervention (CHW) = 164 Control (CHW) = 164 Intervention (CDA) = 249 Control (CDA) = 249 | Intervention (CHW) = 145 Control (CHW) = 149 Intervention (CDA) = 211 Control (CDA) = 215 | 9–17 | 12 |
| Galasso et al,34 Madagascar | Cluster RCT 25 regions per trial arm | • Rural regions • 5-arm trial: control (T0), intensive nutrition counseling (T1), T1+ lipid-based nutrient supplementation (LNS) to children aged 6–18 mo (T2), T1+ LNS to pregnant women, lactating mothers of children aged 0–5 mo and children aged 6–30 mo (T3) and T1+ stimulation to children aged 6–30 mo (T4) • Intervention delivered by community health workers hired to the project | Yes; national nutrition program | Fortnightly home visits | Intervention (T4) = 750 Control (T0) = 747 | Intervention (T4) = 719 Control (T0) = 732 | 0–11 | 24 |
| Grantham-McGrego et al,8 Jamaica | Individual RCT | • Poor urban neighborhoods in Kingston • 4-arm trial: supplementation only, stimulation only, supplementation + stimulation, and control • Targeted stunted children (height-for-age <–2 SD) • Intervention delivered by community health workers hired for the project | No | Weekly home visits | Intervention (stim only) = 30 Control = 33 | Intervention (stim only) = 30 Control = 33 | 12–24 | 24 |
| Grantham-McGregor et al,13 India | Cluster RCT 48 villages per trial arm | • Rural villages in Cuttack, Salepur and Bolangir • 4-arm trial: control, nutritional education, nutritional education + home visits, nutritional education + group sessions • Intervention delivered by local women hired to the project | No | Weekly home visits or weekly group sessions of 7–8 mother–child pairs | HVs + nutr. edu. = 357 Groups + nutr. edu. = 346 Control = 353 | HVs + nutr. edu. = 332 Groups + nutr. edu. = 323 Control = 320 | 7–16 | 24 |
| Hamadani et al,26 Bangladesh | Cluster RCT 10 community nutrition centers per trial arm | • Poor rural area of Monohardi subdistrict • Targeted children with moderate and severe undernutrition (wt for age <–2 SD) • Intervention delivered by local women hired to the project | No | Group meetings: weekly for 10 mths and fortnightly for two mths Home visits: twice weekly for 8 mths and weekly for 4 mths | Intervention = 104 Control = 102 | Intervention = 92 Control =101 | 6–24 | 12 |
| Hamadani et al,12 Bangladesh | Cluster RCT 45 community clinics per trial arm | • Rural sub-districts in Narsingdi • Targeted underweight children (wt for age < -2 SD) • Intervention delivered by government health workers at community clinics | Yes | Group sessions (pairs of mothers) | Intervention = 358 Control = 360 | Intervention = 343 Control = 344 | 5–24 | 12 |
| Heckman et al,29 China | Cluster RCT 55 villages per trial arm | • Rural villages in Hauchi County • Targets disadvantaged families • Intervention delivered by home visitors hired by the project | No | Weekly home visits | Intervention =715 Control = 852 | Intervention = 541 Control = 547 | 0–24 | 22 |
| Hossain et al,35 Bangladesh | Cluster RCT 11 areas per trial arm | • Poor rural sub-district of Ullapara • 3-arm trial: psychosocial stimulation (PS) + unconditional cash transfer (UCT), UCT only and standard care group (comparison). • Intervention delivered by female village health workers hired by project | Yes; Government maternity allowance program | Fortnightly home visits | PS+UCT = 197 UCT only = 188 | PS+UCT = 182 UCT only = 179 | 6–16 | 12 |
| Mehrin et al,14 Bangladesh | Cluster RCT 20 community clinics per trial arm | • Rural district of Kishorganji • Children with weight for age < –1.5 SD, not hospitalized or requiring constant monitoring, and lived within 30-min walk from clinic • Intervention delivered by government health workers at community clinics | Yes | Fortnightly group sessions Groups of 4 mother– child pairs | Intervention = 419 Control = 366 | Intervention = 396 Control = 319 | 5–23 | 12 |
| Nahar et al,27 Bangladesh | Individual RCT | • Urban slums in Dhaka city • 4-arm trial: psychosocial (PS), food supplementation (FS), psychosocial + food stimulation (PS + FS), clinic control (CC) and hospital control (CH) • Targeted severely malnourished children (weight for age <–3 SD) without acute infections or requiring hospitalization • Intervention delivered by local women hired for the project | No | Fortnightly One or more mother–child pair at community clinic | Intervention (PS) = 102 Control (CC) = 99 | Intervention (PS) = 59 Control (CC) = 59 | 6–24 | 6 |
| Powell et al,36 Jamaica | Individual RCT | • Poor urban area in Kingston • Intervention delivered by community health workers from neighborhood clinic • 2 studies are reported; however, only study 2 (an RCT) is included in this evaluation | Yes | Weekly home visits | Intervention = 29 Control = 29 | Intervention = 29 Control = 29 | 16–30 | 12 |
| Powell et al,37 Jamaica | Cluster RCT 11 (intervention) and 7 (control) nutrition clinics | • Urban areas of Kingston and St Andrew • Targeted undernourished children weight for age <–1.5 SD and <–2 SD in the past 3 mo • Intervention delivered by community health aides from primary health nutrition clinics | Yes | Weekly home visits | Intervention = 70 Control = 69 | Intervention = 65 Control = 64 | 9–30 | 12 |
| Tofail et al,38 Bangladesh | Cluster RCT 15 villages per trial arm | • Rural villages in Monohordi subdistrict. • Parallel trials: children who had iron deficiency anemia (IDA) and children neither anemic nor iron deficient (NANI) • In addition to the psychosocial stimulation, iron syrup was given to children with IDA for the first 6 mo • Intervention delivered by local women hired to the project | No | Weekly home visits | Intervention (IDA) = 117 Control (IDA) = 108 Intervention (NANI) = 106 Control (NANI) = 103 | Intervention (IDA) = 110 Control (IDA) = 106 Intervention (NANI) = 104 Control (NANI) = 92 | 6–24 | 9 |
| Walker et al,17 Jamaica | Individual RCT | • Urban areas of Kingston and St Andrew • Targeted term low-birth-weight infants, gestational age ≥37 wk, birth weight <2500 g • Intervention: phase 1: first 8 weeks after birth; phase 2: from age 7 mo to 24 mo • Intervention delivered by community health workers hired for the project | No | Weekly home visits | Intervention = 70 Control = 70 | Intervention = 63 Control = 67 | Birth | 19 |
| Walker et al,39 Jamaica | Cluster RCT 5 centers per trial arm | • Children attending health centers in urban areas of Kingston and St Andrew • 4 trial arms: health center, home visits, health center + home visits, and control • Delivered by health center community health workers | Yes | Fortnightly home visits | Intervention (home visits) = 50 Control = 150 | Intervention (home visits) = 38 Control = 123 | 6 | 12 |
Group-specific analytic sample sizes were not in the published manuscript for all outcomes. In these cases, the study flow diagram was used to approximate sample size.
Primary and Secondary Outcomes Across Studies
| Author, Country | Primary Outcomes | Secondary Outcomes | |||
|---|---|---|---|---|---|
| Cognition | Language | Motor | Home Stimulation | Maternal Depressive Symptoms | |
| Andrew et al,25 India | Bayley-IIIa | Bayley-III | Bayley-III | FCIb | CES-D-6c |
| Araujo et al,21 Peru | ASQ-3d | ASQ-3 | ASQ-3 | FCI and HOMEe | |
| Attanasio et al,23 Colombia | Bayley-III | Bayley-III | Bayley-III | FCI | CES-D-10 |
| Brentani et al,31 Brazil | PRIDIf | PRIDI | PRIDI | FCI | EPDSg |
| Galasso et al,34 Madagascar | ASQ-I | ASQ-I | ASQ-I | FCI | |
| Grantham-McGregor et al,8 Jamaica | Griffithsh | Griffiths | Griffiths | HOME | |
| Grantham-McGregor et al,13 India | Bayley-III | Bayley-III | Bayley-III | FCI and IT-HOMEi | |
| Hamadani et al,26 Bangladesh | Bayley-II | Bayley-II | Bayley-II | ||
| Hamadani et al,12 Bangladesh | Bayley-III | Bayley-III | Bayley-III | FCI | CES-D-6 |
| Heckman et al,29 China | Denverj-II | Denver-II | Denver-II | IT-HOME | |
| Hossain et al,35 Bangladesh | Bayley-III | Bayley-III | Bayley-III | FCI | WHO-SRQ-20k |
| Mehrin et al,14 Bangladesh | Bayley-III | Bayley-III | Bayley-III | FCI | CES-D-6 |
| Nahar et al,27 Bangladesh | Bayley-II | Bayley-II | Bayley-II | HOME | CES-Dl |
| Powell et al,36 Jamaica | Griffiths | Griffiths | Griffiths | CES-D | |
| Powell et al,37 Jamaica | Griffiths | Griffiths | Griffiths | HOME | CES-D-20 |
| Tofail et al,38 Bangladesh | Bayley-II | Bayley-II | Bayley-II | FCI | |
| Walker et al,17 Jamaica | Griffiths | Griffiths | Griffiths | HOME | |
| Walker et al,39 Jamaica | Griffiths | Griffiths | Griffiths | HOME | CES-D |
| Author, Country | Primary Outcomes | Secondary Outcomes | |||
|---|---|---|---|---|---|
| Cognition | Language | Motor | Home Stimulation | Maternal Depressive Symptoms | |
| Andrew et al,25 India | Bayley-IIIa | Bayley-III | Bayley-III | FCIb | CES-D-6c |
| Araujo et al,21 Peru | ASQ-3d | ASQ-3 | ASQ-3 | FCI and HOMEe | |
| Attanasio et al,23 Colombia | Bayley-III | Bayley-III | Bayley-III | FCI | CES-D-10 |
| Brentani et al,31 Brazil | PRIDIf | PRIDI | PRIDI | FCI | EPDSg |
| Galasso et al,34 Madagascar | ASQ-I | ASQ-I | ASQ-I | FCI | |
| Grantham-McGregor et al,8 Jamaica | Griffithsh | Griffiths | Griffiths | HOME | |
| Grantham-McGregor et al,13 India | Bayley-III | Bayley-III | Bayley-III | FCI and IT-HOMEi | |
| Hamadani et al,26 Bangladesh | Bayley-II | Bayley-II | Bayley-II | ||
| Hamadani et al,12 Bangladesh | Bayley-III | Bayley-III | Bayley-III | FCI | CES-D-6 |
| Heckman et al,29 China | Denverj-II | Denver-II | Denver-II | IT-HOME | |
| Hossain et al,35 Bangladesh | Bayley-III | Bayley-III | Bayley-III | FCI | WHO-SRQ-20k |
| Mehrin et al,14 Bangladesh | Bayley-III | Bayley-III | Bayley-III | FCI | CES-D-6 |
| Nahar et al,27 Bangladesh | Bayley-II | Bayley-II | Bayley-II | HOME | CES-Dl |
| Powell et al,36 Jamaica | Griffiths | Griffiths | Griffiths | CES-D | |
| Powell et al,37 Jamaica | Griffiths | Griffiths | Griffiths | HOME | CES-D-20 |
| Tofail et al,38 Bangladesh | Bayley-II | Bayley-II | Bayley-II | FCI | |
| Walker et al,17 Jamaica | Griffiths | Griffiths | Griffiths | HOME | |
| Walker et al,39 Jamaica | Griffiths | Griffiths | Griffiths | HOME | CES-D |
Bayley Scales of Infant and Toddler Development.
UNICEF’s Family Care Indicators.
Center for Epidemiologic Studies Depression Scale, 6 items.
Ages and Stages Questionnaire.
Home Observation for Measurement of the Environment.
Regional Project on Child Development Indicators (specifically, Eagle’s Scales of Child Development).
Edinburgh Postnatal Depression Scale.
Griffiths Scales of Mental Development.
Infant-Toddler Home Observation for Measurement of the Environment.
Denver Developmental Screening Test.
Validated Bengali version of World Health Organization Self-reported Questionnaire-20.
Center for Epidemiologic Studies Depression Scale, adapted version used by Baker-Henningham et al (2005).
Figure 2 shows the overall effect size for each outcome using the Hedges’ g SMD, and the effect sizes in SDs of the instruments used for each study and the percentages of total weight for each study, with the weights representing study precision. For cognition (panel A), the overall effect size is 0.49 (confidence interval [CI], 0.32 to 0.66; P < .00; I2 = 90.46%) for 14 trials. Adding trials that report cognition alone to those that report combined cognition and language, the overall effect size is 0.39 (CI, 0.24 to 0.53; P < .00; I2 = 89.70%) and the number of trials is 20. For language (panel C), the overall effect size is 0.38 (CI, 0.24 to 0.51; P < .00; I2 = 87.12%) for 15 trials. Panel D shows analysis adding trials that report language alone with those that report combined cognition and language with an overall effect size of 0.31 (CI, 0.20 to 0.43; P < .00; I2 = 85.27%) for 21 trials. Eight trials assessed fine motor (panel E) and the overall effect size is 0.29 (CI, 0.12 to 0.47; P < .00; I2 = 75.44%). The overall effect size for gross motor for six trials is 0.05 (CI -0.05 to 0.15; P < .11; I2 = 33.33%) (panel F). Adding trials that report fine motor alone to those that report combined fine and gross motor (panel G), the overall effect size is 0.27 (CI, 0.13 to 0.40; P < .00; I2 = 87.96%) for 19 trials.
Forest plots for the effect of interventions on child outcomes, home stimulation, and maternal depressive symptoms. SMD, standardized mean difference. Blue squares represent the SMD for each trial, with the size of the square being proportional to the trial weight. The whiskers extending from each side of the square represent the range of 95% CI. The green diamond shows the overall pooled effect size using a random-effect model, which is centered at the point estimate and the diamond width representing the 95% CI. Data and other materials used in this study are available on request.
Forest plots for the effect of interventions on child outcomes, home stimulation, and maternal depressive symptoms. SMD, standardized mean difference. Blue squares represent the SMD for each trial, with the size of the square being proportional to the trial weight. The whiskers extending from each side of the square represent the range of 95% CI. The green diamond shows the overall pooled effect size using a random-effect model, which is centered at the point estimate and the diamond width representing the 95% CI. Data and other materials used in this study are available on request.
The effect sizes for the secondary outcomes, home stimulation (panel H, 17 trials) and maternal depressive symptoms (panel I, 12 trials), are 0.37 (CI, 0.21 to 0.54; P < .00; I2 = 92.79%) and –0.09 (CI, –0.19 to 0.01; P < .00; I2 = 62.61%), respectively.
Table 3 shows the heterogeneity analysis stratified by maternal education at the individual level using 2 different classifications of education. There were no significant differences in effects on child outcomes, home stimulation, and maternal depressive symptoms when using subgroups defined by either education classification.
Subgroup Results for the Effect of Interventions on Child Outcomes, Home Stimulation, and Maternal Depressive Symptoms Stratified by Maternal Education
| Cognition | Cognition and Combined Cognition and Language | Language | Language and Combined Cognition and Language | Fine Motor and Combined Fine and Gross Motor | Home Stimulation | Maternal Depressive Symptoms | |||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Moderator | N | SMD (95% CI) | P | N | SMD (95% CI) | P | N | SMD (95% CI) | P | N | SMD (95% CI) | P | N | SMD (95% CI) | P | N | SMD (95% CI) | P | N | SMD (95% CI) | P |
| Definition 1 | |||||||||||||||||||||
| Primary education or less | 8 | 0.55 (0.15–0.94) | .98 | 14 | 0.37 (0.12–0.62) | .75 | 9 | 0.39 (0.14–0.64) | .97 | 15 | 0.29 (0.12–0.46) | .65 | 14 | 0.18 (–0.03 to 0.38) | .56 | 13 | 0.38 (0.22–0.55) | .40 | 10 | −0.09 (–0.18 to 0.00) | .65 |
| More than primary education | 8 | 0.54 (0.11–0.96) | 14 | 0.31 (0.01–0.60) | 9 | 0.39 (0.14–0.65) | 15 | 0.23 (0.03–0.43) | 14 | 0.27 (0.03–0.51) | 13 | 0.28 (0.10–0.46) | 10 | −0.05 (–0.20 to 0.09) | |||||||
| Definition 2 | |||||||||||||||||||||
| Lower half education | 10 | 0.58 (0.27–0.88) | .91 | 16 | 0.41 (0.19–0.63) | .89 | 11 | 0.44 (0.21–0.66) | .73 | 17 | 0.33 (0.17–0.50) | .70 | 16 | 0.28 (0.09–0.46) | .90 | 15 | 0.40 (0.24–0.55) | .35 | 11 | −0.11 (–0.20 to −0.01) | .46 |
| Upper half education | 10 | 0.55 (0.18–0.92) | 16 | 0.39 (0.14–0.64) | 11 | 0.38 (0.13–0.63) | 17 | 0.29 (0.11–0.46) | 16 | 0.26 (0.02–0.49) | 15 | 0.29 (0.14–0.44) | 11 | −0.04 (–0.18 to 0.09) | |||||||
| Cognition | Cognition and Combined Cognition and Language | Language | Language and Combined Cognition and Language | Fine Motor and Combined Fine and Gross Motor | Home Stimulation | Maternal Depressive Symptoms | |||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Moderator | N | SMD (95% CI) | P | N | SMD (95% CI) | P | N | SMD (95% CI) | P | N | SMD (95% CI) | P | N | SMD (95% CI) | P | N | SMD (95% CI) | P | N | SMD (95% CI) | P |
| Definition 1 | |||||||||||||||||||||
| Primary education or less | 8 | 0.55 (0.15–0.94) | .98 | 14 | 0.37 (0.12–0.62) | .75 | 9 | 0.39 (0.14–0.64) | .97 | 15 | 0.29 (0.12–0.46) | .65 | 14 | 0.18 (–0.03 to 0.38) | .56 | 13 | 0.38 (0.22–0.55) | .40 | 10 | −0.09 (–0.18 to 0.00) | .65 |
| More than primary education | 8 | 0.54 (0.11–0.96) | 14 | 0.31 (0.01–0.60) | 9 | 0.39 (0.14–0.65) | 15 | 0.23 (0.03–0.43) | 14 | 0.27 (0.03–0.51) | 13 | 0.28 (0.10–0.46) | 10 | −0.05 (–0.20 to 0.09) | |||||||
| Definition 2 | |||||||||||||||||||||
| Lower half education | 10 | 0.58 (0.27–0.88) | .91 | 16 | 0.41 (0.19–0.63) | .89 | 11 | 0.44 (0.21–0.66) | .73 | 17 | 0.33 (0.17–0.50) | .70 | 16 | 0.28 (0.09–0.46) | .90 | 15 | 0.40 (0.24–0.55) | .35 | 11 | −0.11 (–0.20 to −0.01) | .46 |
| Upper half education | 10 | 0.55 (0.18–0.92) | 16 | 0.39 (0.14–0.64) | 11 | 0.38 (0.13–0.63) | 17 | 0.29 (0.11–0.46) | 16 | 0.26 (0.02–0.49) | 15 | 0.29 (0.14–0.44) | 11 | −0.04 (–0.18 to 0.09) | |||||||
N = number of trials represented in subgroup analysis. P value corresponds to test of subgroup differences.
Table 4 shows the study-level heterogeneity by the 9 possible moderator variables described in the analysis section. We do not include fine motor and gross motor separately because there are only 8 and 6 trials, respectively, leading to some subgroups with <3 trials. Average effect sizes were larger in subgroups with mean child age on enrollment >12 months for cognition, language, and combined motor development, and larger in studies targeted to undernourished children in all domains.
Subgroup Results for the Effect of Interventions on Child Outcomes, Home Stimulation, and Maternal Depressive Symptoms Stratified by Trial Level Characteristics
| Cognition | Cognition and Combined Cognition and Language | Language | Language and Combined Cognition and Language | Fine Motor and combined Fine and Gross Motor | Home Stimulation | Maternal Depressive Symptoms | |||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Moderator | N | SMD (95% CI) | P | N | SMD (95% CI) | P | N | SMD (95% CI) | P | N | SMD (95% CI) | P | N | SMD (95% CI) | P | N | SMD (95% CI) | P | N | SMD (95% CI) | P |
| Trials stratified by enrollment age | |||||||||||||||||||||
| Mean age ≤12 mo | 7 | 0.29 (0.16 to 0.43) | .01* | 8 | 0.29 (0.17 to 0.42) | .26 | 7 | 0.23 (0.11 to 0.34) | .02* | 8 | 0.23 (0.12 to 0.34) | .20 | 8 | 0.10 (0.04 to 0.17) | .02* | 8 | 0.34 (0.18 to 0.50) | .71 | 5 | −0.02 (–0.12 to 0.08) | .27 |
| >12 mo | 7 | 0.67 (0.40 to 0.94) | 12 | 0.44 (0.21 to 0.66) | 8 | 0.52 (0.31 to 0.73) | 13 | 0.37 (0.19 to 0.54) | 11 | 0.36 (0.16 to 0.57) | 9 | 0.40 (0.12 to 0.68) | 7 | −0.12 (–0.27 to 0.03) | |||||||
| Trials stratified by child targeted to undernourished | |||||||||||||||||||||
| No | 9 | 0.29 (0.18 to 0.39) | .00* | 13 | 0.23 (0.12 to 0.33) | .00* | 10 | 0.25 (0.16 to 0.34) | .04* | 14 | 0.21 (0.12 to 0.30) | .02* | 12 | 0.11 (0.06 to 0.17) | .01* | 11 | 0.23 (0.09 to 0.37) | .00* | 8 | −0.02 (-0.11 to 0.08) | .00* |
| Yes | 5 | 0.83 (0.59 to 1.07) | 7 | 0.67 (0.42 to 0.93) | 5 | 0.58 (0.28 to 0.87) | 7 | 0.50 (0.28 to 0.73) | 7 | 0.49 (0.21 to 0.77) | 6 | 0.67 (0.40 to 0.94) | 4 | −0.24 (–0.37 to −0.12) | |||||||
| Trials stratified by number of children receiving intervention | |||||||||||||||||||||
| <300 | 8 | 0.60 (0.37 to 0.82) | .17 | 13 | 0.41 (0.21 to 0.60) | .75 | 8 | 0.48 (0.25 to 0.72) | .14 | 13 | 0.34 (0.16 to 0.52) | .62 | 11 | 0.34 (0.12 to 0.55) | .23 | 10 | 0.41 (0.18 to 0.64) | .63 | 8 | −0.13 (–0.27 to 0.02) | .33 |
| ≥300 | 6 | 0.36 (0.12 to 0.61) | 7 | 0.36 (0.15 to 0.57) | 7 | 0.28 (0.14 to 0.42) | 8 | 0.28 (0.16 to 0.41) | 8 | 0.18 (0.07 to 0.30) | 7 | 0.33 (0.09 to 0.57) | 4 | −0.03 (–0.14 to 0.07) | |||||||
| Trials stratified by type of delivery | |||||||||||||||||||||
| HV | 11 | 0.36 (0.23 to 0.49) | .07* | 17 | 0.28 (0.17 to 0.39) | .03* | 12 | 0.28 (0.17 to 0.40) | .03* | 18 | 0.23 (0.14 to 0.33) | .01* | 16 | 0.15 (0.08 to 0.21) | .09* | 14 | 0.27 (0.14 to 0.40) | .05* | 9 | −0.06 (–0.17 to 0.05) | .52 |
| Group | 3 | 0.79 (0.34 to 1.24) | 3 | 0.79 (0.34 to 1.24) | 3 | 0.61 (0.34 to 0.88) | 3 | 0.61 (0.34 to 0.88) | 3 | 0.58 (0.09 to 1.06) | 3 | 0.77 (0.29 to 1.25) | 3 | −0.14 (–0.37 to 0.09) | |||||||
| Trials stratified by the duration of the intervention | |||||||||||||||||||||
| ≤12 mo | 6 | 0.73 (0.47 to 0.99) | .00* | 12 | 0.44 (0.22 to 0.66) | .22 | 6 | 0.60 (0.39 to 0.82) | .00* | 12 | 0.37 (0.19 to 0.56) | .14 | 10 | 0.35 (0.12 to 0.57) | .06* | 8 | 0.52 (0.20 to 0.83) | .07* | 8 | −0.12 (–0.26 to 0.02) | .38 |
| >12 mo | 8 | 0.28 (0.16 to 0.40) | 8 | 0.28 (0.16 to 0.40) | 9 | 0.22 (0.13 to 0.31) | 9 | 0.22 (0.13 to 0.31) | 9 | 0.12 (0.06 to 0.19) | 9 | 0.22 (0.14 to 0.29) | 4 | −0.04 (–0.16 to 0.09) | |||||||
| Trials stratified by the frequency of visits | |||||||||||||||||||||
| Fortnightly | 6 | 0.52 (0.18 to 0.87) | .52 | 9 | 0.37 (0.09 to 0.65) | .90 | 6 | 0.39 (0.14 to 0.64) | .72 | 9 | 0.28 (0.07 to 0.49) | .86 | 7 | 0.31 (0.02 to 0.60) | .43 | 9 | 0.42 (0.13 to 0.71) | .39 | 7 | −0.09 (–0.23 to 0.05) | .90 |
| Weekly | 8 | 0.41 (0.28 to 0.53) | 11 | 0.35 (0.27 to 0.44) | 9 | 0.34 (0.20 to 0.48) | 12 | 0.30 (0.21 to 0.39) | 12 | 0.19 (0.10 to 0.27) | 8 | 0.28 (0.18 to 0.39) | 5 | −0.08 (–0.22 to 0.06) | |||||||
| Trials stratified by the frequency of supervision (observations) | |||||||||||||||||||||
| Weekly or fortnightly | 4 | 0.50 (0.22 to 0.77) | .94 | 4 | 0.50 (0.22 to 0.77) | .38 | 4 | 0.40 (0.26 to 0.55) | .81 | 4 | 0.40 (0.26 to 0.55) | .27 | 4 | 0.21 (–0.01 to 0.42) | .56 | 4 | 0.48 (0.11 to 0.85) | .50 | 4 | −0.08 (–0.22 to 0.07) | .87 |
| Monthly or less often | 10 | 0.48 (0.26 to 0.71) | 16 | 0.36 (0.19 to 0.52) | 11 | 0.37 (0.18 to 0.56) | 17 | 0.29 (0.15 to 0.43) | 15 | 0.29 (0.12 to 0.45) | 13 | 0.34 (0.15 to 0.53) | 8 | −0.09 (–0.24 to 0.05) | |||||||
| Trials stratified by who pay home visitors/group facilitators | |||||||||||||||||||||
| Paid by government | 7 | 0.33 (0.25 to 0.41) | .09* | 12 | 0.29 (0.21 to 0.37) | .19 | 8 | 0.27 (0.20 to 0.34) | .09* | 13 | 0.25 (0.18 to 0.32) | .23 | 12 | 0.15 (0.09 to 0.21) | .04* | 10 | 0.29 (0.13 to 0.45) | .31 | 7 | −0.04 (–0.12 to 0.03) | .37 |
| Paid by project | 7 | 0.61 (0.30 to 0.92) | 8 | 0.51 (0.19 to 0.83) | 7 | 0.50 (0.24 to 0.77) | 8 | 0.42 (0.15 to 0.70) | 7 | 0.45 (0.17 to 0.74) | 7 | 0.48 (0.16 to 0.80) | 5 | −0.15 (–0.38 to 0.07) | |||||||
| Trials stratified by geographic location | |||||||||||||||||||||
| Rural | 8 | 0.46 (0.20 to 0.72) | .76 | 11 | 0.41 (0.21 to 0.61) | .69 | 9 | 0.35 (0.19 to 0.52) | .65 | 12 | 0.33 (0.20 to 0.47) | .74 | 12 | 0.25 (0.09 to 0.42) | .80 | 9 | 0.45 (0.20 to 0.70) | .20 | 6 | −0.10 (–0.22 to 0.02) | .91 |
| Urban | 6 | 0.51 (0.33 to 0.69) | 9 | 0.35 (0.14 to 0.56) | 6 | 0.43 (0.16 to 0.71) | 9 | 0.29 (0.07 to 0.51) | 7 | 0.29 (0.05 to 0.53) | 8 | 0.25 (0.07 to 0.44) | 6 | −0.08 (–0.27 to 0.10) | |||||||
| Cognition | Cognition and Combined Cognition and Language | Language | Language and Combined Cognition and Language | Fine Motor and combined Fine and Gross Motor | Home Stimulation | Maternal Depressive Symptoms | |||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Moderator | N | SMD (95% CI) | P | N | SMD (95% CI) | P | N | SMD (95% CI) | P | N | SMD (95% CI) | P | N | SMD (95% CI) | P | N | SMD (95% CI) | P | N | SMD (95% CI) | P |
| Trials stratified by enrollment age | |||||||||||||||||||||
| Mean age ≤12 mo | 7 | 0.29 (0.16 to 0.43) | .01* | 8 | 0.29 (0.17 to 0.42) | .26 | 7 | 0.23 (0.11 to 0.34) | .02* | 8 | 0.23 (0.12 to 0.34) | .20 | 8 | 0.10 (0.04 to 0.17) | .02* | 8 | 0.34 (0.18 to 0.50) | .71 | 5 | −0.02 (–0.12 to 0.08) | .27 |
| >12 mo | 7 | 0.67 (0.40 to 0.94) | 12 | 0.44 (0.21 to 0.66) | 8 | 0.52 (0.31 to 0.73) | 13 | 0.37 (0.19 to 0.54) | 11 | 0.36 (0.16 to 0.57) | 9 | 0.40 (0.12 to 0.68) | 7 | −0.12 (–0.27 to 0.03) | |||||||
| Trials stratified by child targeted to undernourished | |||||||||||||||||||||
| No | 9 | 0.29 (0.18 to 0.39) | .00* | 13 | 0.23 (0.12 to 0.33) | .00* | 10 | 0.25 (0.16 to 0.34) | .04* | 14 | 0.21 (0.12 to 0.30) | .02* | 12 | 0.11 (0.06 to 0.17) | .01* | 11 | 0.23 (0.09 to 0.37) | .00* | 8 | −0.02 (-0.11 to 0.08) | .00* |
| Yes | 5 | 0.83 (0.59 to 1.07) | 7 | 0.67 (0.42 to 0.93) | 5 | 0.58 (0.28 to 0.87) | 7 | 0.50 (0.28 to 0.73) | 7 | 0.49 (0.21 to 0.77) | 6 | 0.67 (0.40 to 0.94) | 4 | −0.24 (–0.37 to −0.12) | |||||||
| Trials stratified by number of children receiving intervention | |||||||||||||||||||||
| <300 | 8 | 0.60 (0.37 to 0.82) | .17 | 13 | 0.41 (0.21 to 0.60) | .75 | 8 | 0.48 (0.25 to 0.72) | .14 | 13 | 0.34 (0.16 to 0.52) | .62 | 11 | 0.34 (0.12 to 0.55) | .23 | 10 | 0.41 (0.18 to 0.64) | .63 | 8 | −0.13 (–0.27 to 0.02) | .33 |
| ≥300 | 6 | 0.36 (0.12 to 0.61) | 7 | 0.36 (0.15 to 0.57) | 7 | 0.28 (0.14 to 0.42) | 8 | 0.28 (0.16 to 0.41) | 8 | 0.18 (0.07 to 0.30) | 7 | 0.33 (0.09 to 0.57) | 4 | −0.03 (–0.14 to 0.07) | |||||||
| Trials stratified by type of delivery | |||||||||||||||||||||
| HV | 11 | 0.36 (0.23 to 0.49) | .07* | 17 | 0.28 (0.17 to 0.39) | .03* | 12 | 0.28 (0.17 to 0.40) | .03* | 18 | 0.23 (0.14 to 0.33) | .01* | 16 | 0.15 (0.08 to 0.21) | .09* | 14 | 0.27 (0.14 to 0.40) | .05* | 9 | −0.06 (–0.17 to 0.05) | .52 |
| Group | 3 | 0.79 (0.34 to 1.24) | 3 | 0.79 (0.34 to 1.24) | 3 | 0.61 (0.34 to 0.88) | 3 | 0.61 (0.34 to 0.88) | 3 | 0.58 (0.09 to 1.06) | 3 | 0.77 (0.29 to 1.25) | 3 | −0.14 (–0.37 to 0.09) | |||||||
| Trials stratified by the duration of the intervention | |||||||||||||||||||||
| ≤12 mo | 6 | 0.73 (0.47 to 0.99) | .00* | 12 | 0.44 (0.22 to 0.66) | .22 | 6 | 0.60 (0.39 to 0.82) | .00* | 12 | 0.37 (0.19 to 0.56) | .14 | 10 | 0.35 (0.12 to 0.57) | .06* | 8 | 0.52 (0.20 to 0.83) | .07* | 8 | −0.12 (–0.26 to 0.02) | .38 |
| >12 mo | 8 | 0.28 (0.16 to 0.40) | 8 | 0.28 (0.16 to 0.40) | 9 | 0.22 (0.13 to 0.31) | 9 | 0.22 (0.13 to 0.31) | 9 | 0.12 (0.06 to 0.19) | 9 | 0.22 (0.14 to 0.29) | 4 | −0.04 (–0.16 to 0.09) | |||||||
| Trials stratified by the frequency of visits | |||||||||||||||||||||
| Fortnightly | 6 | 0.52 (0.18 to 0.87) | .52 | 9 | 0.37 (0.09 to 0.65) | .90 | 6 | 0.39 (0.14 to 0.64) | .72 | 9 | 0.28 (0.07 to 0.49) | .86 | 7 | 0.31 (0.02 to 0.60) | .43 | 9 | 0.42 (0.13 to 0.71) | .39 | 7 | −0.09 (–0.23 to 0.05) | .90 |
| Weekly | 8 | 0.41 (0.28 to 0.53) | 11 | 0.35 (0.27 to 0.44) | 9 | 0.34 (0.20 to 0.48) | 12 | 0.30 (0.21 to 0.39) | 12 | 0.19 (0.10 to 0.27) | 8 | 0.28 (0.18 to 0.39) | 5 | −0.08 (–0.22 to 0.06) | |||||||
| Trials stratified by the frequency of supervision (observations) | |||||||||||||||||||||
| Weekly or fortnightly | 4 | 0.50 (0.22 to 0.77) | .94 | 4 | 0.50 (0.22 to 0.77) | .38 | 4 | 0.40 (0.26 to 0.55) | .81 | 4 | 0.40 (0.26 to 0.55) | .27 | 4 | 0.21 (–0.01 to 0.42) | .56 | 4 | 0.48 (0.11 to 0.85) | .50 | 4 | −0.08 (–0.22 to 0.07) | .87 |
| Monthly or less often | 10 | 0.48 (0.26 to 0.71) | 16 | 0.36 (0.19 to 0.52) | 11 | 0.37 (0.18 to 0.56) | 17 | 0.29 (0.15 to 0.43) | 15 | 0.29 (0.12 to 0.45) | 13 | 0.34 (0.15 to 0.53) | 8 | −0.09 (–0.24 to 0.05) | |||||||
| Trials stratified by who pay home visitors/group facilitators | |||||||||||||||||||||
| Paid by government | 7 | 0.33 (0.25 to 0.41) | .09* | 12 | 0.29 (0.21 to 0.37) | .19 | 8 | 0.27 (0.20 to 0.34) | .09* | 13 | 0.25 (0.18 to 0.32) | .23 | 12 | 0.15 (0.09 to 0.21) | .04* | 10 | 0.29 (0.13 to 0.45) | .31 | 7 | −0.04 (–0.12 to 0.03) | .37 |
| Paid by project | 7 | 0.61 (0.30 to 0.92) | 8 | 0.51 (0.19 to 0.83) | 7 | 0.50 (0.24 to 0.77) | 8 | 0.42 (0.15 to 0.70) | 7 | 0.45 (0.17 to 0.74) | 7 | 0.48 (0.16 to 0.80) | 5 | −0.15 (–0.38 to 0.07) | |||||||
| Trials stratified by geographic location | |||||||||||||||||||||
| Rural | 8 | 0.46 (0.20 to 0.72) | .76 | 11 | 0.41 (0.21 to 0.61) | .69 | 9 | 0.35 (0.19 to 0.52) | .65 | 12 | 0.33 (0.20 to 0.47) | .74 | 12 | 0.25 (0.09 to 0.42) | .80 | 9 | 0.45 (0.20 to 0.70) | .20 | 6 | −0.10 (–0.22 to 0.02) | .91 |
| Urban | 6 | 0.51 (0.33 to 0.69) | 9 | 0.35 (0.14 to 0.56) | 6 | 0.43 (0.16 to 0.71) | 9 | 0.29 (0.07 to 0.51) | 7 | 0.29 (0.05 to 0.53) | 8 | 0.25 (0.07 to 0.44) | 6 | −0.08 (–0.27 to 0.10) | |||||||
N = number of studies represented in subgroup analysis. P value corresponds to test of subgroup differences.
Indicates significant moderator effect (P < .1).
We identified multiple variations in the implementation of the Reach Up program across the 18 studies for which we defined 9 types of stratification, specifically, by enrollment age of children, child targeted to undernourished, number of children receiving intervention, type of delivery, frequency of visits, frequency of supervision, who pay the program staff, and geographic location. These implementation methods were also significant moderators. Delivery by groups showed significantly larger effect sizes compared with home visits for cognition and combined cognition and language, language, and language and combined cognition and language and P < .07 for cognition. Trials with duration ≤12 months showed larger effect sizes than those with longer duration for cognition, language, and motor development. Where delivery staff were paid by a project, there was a tendency for effect sizes to be larger, which was significant for motor development and P < .09 for cognition and language.
The only moderators that significantly affected impact on home stimulation were undernutrition and group delivery and targeting undernutrition was the only moderator that affected maternal depressive symptoms, reducing it for mothers of undernourished children.
For the remaining four variables (number of children receiving intervention, frequency of visits, frequency of supervision, geographic location), there was no significant heterogeneity.
Six of the 18 studies had a low overall risk of bias and 12 were rated as “some concerns” (Supplemental Table 7). The domains that included some concerns were selection bias (4/18 studies), performance bias (4/18 studies), and reporting bias (11/18). Risk of reporting bias was due to lack of a preanalysis plan for these studies (Supplemental Table 8). The GRADE assessment for the primary outcomes showed that heterogeneity was rated low on 5 outcomes (“cognition,” “cognition and combined cognition and language,” “language,” “language and combined cognition and language,” and “fine motor and combined fine and gross motor”) and imprecision was rated high for 3 outcomes (“language,” “language and combined cognition and language,” and “fine motor and combined fine and gross motor”). All outcomes were rated high for indirectness and moderate for risk of bias and publication bias. The overall quality of the evidence for all outcomes was moderate (Table 5).
GRADE Summary of Findings
| Primary Outcomes | Absolute Effect (95% CI) | N Participants (Trials) | Heterogeneity I2 | Quality of Evidence (GRADEa) |
|---|---|---|---|---|
| Cognition | 0.49 (0.32 to 0.66) | 10 401 (14) | 0.90 | Moderate |
| Cognition and combined cognition and language | 0.39 (0.24 to 0.63) | 11 657 (20) | 0.90 | Moderate |
| Language | 0.38 (0.24 to 0.51) | 11 398 (15) | 0.87 | Moderate |
| Language and combined cognition and language | 0.31 (0.20 to 0.42) | 12 551 (21) | 0.85 | Moderate |
| Fine motor | 0.29 (0.12 to 0.54) | 7682 (8) | 0.75 | High |
| Gross motor | 0.05 (–0.08 to 0.31) | 7463 (6) | 0.33 | High |
| Fine motor and combined fine and gross motor | 0.27 (0.12 to 0.39) | 12 130 (19) | 0.88 | Moderate |
| Overall rating: moderate |
| Primary Outcomes | Absolute Effect (95% CI) | N Participants (Trials) | Heterogeneity I2 | Quality of Evidence (GRADEa) |
|---|---|---|---|---|
| Cognition | 0.49 (0.32 to 0.66) | 10 401 (14) | 0.90 | Moderate |
| Cognition and combined cognition and language | 0.39 (0.24 to 0.63) | 11 657 (20) | 0.90 | Moderate |
| Language | 0.38 (0.24 to 0.51) | 11 398 (15) | 0.87 | Moderate |
| Language and combined cognition and language | 0.31 (0.20 to 0.42) | 12 551 (21) | 0.85 | Moderate |
| Fine motor | 0.29 (0.12 to 0.54) | 7682 (8) | 0.75 | High |
| Gross motor | 0.05 (–0.08 to 0.31) | 7463 (6) | 0.33 | High |
| Fine motor and combined fine and gross motor | 0.27 (0.12 to 0.39) | 12 130 (19) | 0.88 | Moderate |
| Overall rating: moderate |
Grading Recommendations Assessment, Development and Evaluation.
Discussion
The impacts of the Reach Up intervention on child development, in 18 studies across 8 countries, were generally comparable with results from LMIC from 2 recent reviews of ECD programs with the largest benefits on cognition followed by language and then fine motor development.3,6 The average effect size for cognition is somewhat higher, whereas that for language is similar to Jeong et al.6 The earlier reviews report combined fine and gross motor development. We find a slightly larger effect size for studies with fine motor development separately than when combined with studies reporting an overall motor score. The effect size for the combined motor analysis is similar to that in Jeong et al.6 and larger than the earlier review.3
The overall effect size of 0.37 SD for home stimulation was lower than in the 2 previous reviews.5,6 Increased stimulation in the home is an expected mechanism leading to gains in child development. Only 1 of 17 trials reporting on home stimulation examined mediation, finding that increased home stimulation partially mediated gains in development.17 Mediation by stimulation in the home has been demonstrated for a few other parenting interventions in LMIC.18–20 Further evidence on mediation is needed, which, if consistent, may allow use of change in stimulation in the home to monitor implementation at scale.
The small, nonsignificant reduction in maternal depressive symptoms is consistent with previous studies.5,6 This suggests that improving maternal depression will require additional content targeting this objective.
Intervention effects did not vary by maternal education using 2 approaches to define lower and higher education. This suggests the intervention benefits child outcomes and home stimulation across the range of maternal education seen in the studies in this review. This is consistent with previous examination of differences in intervention impact by maternal education in individual studies in this review,12,21,23,26 except 1,22 and with other recent evidence.23,24
Benefits to child outcomes were seen in both younger and older children. However, children with a mean age >12 months on enrollment had larger benefits to cognitive, language, and motor development than younger children. This is important for planning of programs in which it may not be possible to reach all children or the youngest children. A previous meta-analysis also found larger cognitive and language benefits in children older than age 12 months.6 Intervention impact on home stimulation did not vary by child age.
The Reach Up program aims to improve the development of disadvantaged children. The studies reviewed recruited families from poor areas, with some identifying them through social protection programs. Some studies targeted undernourished children (low height or weight for age), and 1 study targeted children born low birth weight at term. We found larger average benefits to child outcomes in these studies (cognition as high as 0.83 SD) and to stimulation in the home, and the undernourished subgroup was the only one to show significant reduction in maternal depressive symptoms. Individual study reports showed larger benefits for children stunted on enrollment in 1 study,25 but effects did not differ by height for age on enrollment in contexts where the prevalence of stunting was high.21,26 These findings have strong policy implications and suggest that targeting undernourished children identifies the most disadvantaged families who benefit more from intervention. Consequently, where resources are limited, targeting should be considered.
There were only 3 studies12–14 with delivery by mother–child groups, and these had higher average effect sizes than home visits delivery. One study randomized mothers and children to home visits or groups of 8 and found similar effects in the 2 delivery methods, with groups being more cost-effective.13 The other 2 studies involved pairs of mothers or groups of 4 and had large effects. In contrast, a significant but small benefit was reported for overall development from group-based delivery of Reach Up,28 no differences by delivery mode were reported in the meta-analysis by Jeong et al,6 and a recent meta-analysis of ECD interventions in China reported greater benefits from home visits than center-based interventions, even though one-on-one as well as group sessions were provided in the centers.30 More studies are needed, but the evidence suggests that the Reach Up intervention remains effective when delivered through groups. Use of groups may be an important strategy for scaling and efforts may be needed to ensure access by the most disadvantaged families and sustain participation.
Interventions in which delivery staff were project-funded tended to have larger effects than those integrated with government services, where staff were paid by the government. This likely reflects the challenges faced as interventions scale because government constraints affect resources for implementation, timely hiring and training of staff, and lack of monitoring of implementation quality. In addition, in most cases in which staff are paid by an existing service, which in most cases is the government, they have many other duties and their workload affects intervention delivery. An extreme case of this is the study in Brazil where home visits by community health workers employed in government primary care centers was attempted. Very few visits were made, with 83% of families receiving no visits.31 In the national program in Peru, 34% of children initially randomized to treatment received no visits because the program never operated in some areas and there was a small increase in effect size to 0.15 SD when those children were excluded from analyses.21
There were larger average effect sizes where intervention duration was 12 months or less compared with longer duration. It is possible that initial impacts are greater, and that gains are maintained but may not increase substantially as programs continue. Most of the shorter duration studies continued for 12 months, with only 1 with duration of 6 months and 1 of 9 months, so the finding does not apply to very short programs. It remains uncertain whether longer duration is important for sustainability of benefits. Sustainability of impact has been examined in 3 studies, all with interventions of 18 to 24 months’ duration. One showed medium-32 and another long-term benefits10 ; however, 1 study did not show benefits at 2-year follow-up.22 There are few follow-up studies of other early stimulation interventions.6 Follow-up of a 2-year ECD intervention delivered by female health workers in Pakistan showed benefits to child cognitive, language, and motor skills 2 years later.33
There were significant benefits to child outcomes and home stimulation and little evidence of heterogeneity in the remaining implementation characteristics, whether visits/group sessions were weekly or fortnightly, in rural or urban areas, whether numbers of children targeted was small (<300) or larger, and if supervision was monthly or less compared with more frequent. For scaling, this provides further evidence that fortnightly contacts can be expected to lead to benefits. Despite lack of significant differences, effect sizes tended to be smaller with increasing numbers of children reached. Furthermore, many of these larger interventions were not at large scale and reached between 350 and 850 children. Nonetheless, this does provide some evidence that significant impact can be attained as programs increase in size. Finally, although effect size did not vary by frequency of supervision, frequency was less than monthly in few studies.
We were not able to examine some other variables that may be important for implementation such as training, as all except 2 studies used recommended training duration, and education level of delivery staff, which was not consistently reported. In 1 study, it was not possible to leave play materials in the homes, which may have contributed to the lack of significant benefits.34
This review has multiple strengths. First, it uses data from intervention trials that followed the same curriculum and were evaluated using RCTs, which facilitates making comparisons. Furthermore, most studies assessed cognitive, motor, and language development with direct assessment. Only 2 studies used the Ages & Stages Questionnaire, relying on both observations and maternal/caregiver’s reports that may be biased if mothers who received the intervention overreport on children’s developmental progress compared with control mothers. However, Araujo et al21 did not detect this bias because larger effects were found for items collected by direct observation (0.13 SD) than by maternal report (0.07 SD). Moreover, we were able to examine several implementation characteristics that may affect intervention outcomes, including targeting, frequency of visits, duration, and integration with government services. Finally, all the multiple levels of trials characteristics provided evidence on how the Reach Up program could be adapted across contexts.
There are strengths and limitations to using a meta-analysis approach in a systematic review. The trials vary in the instruments used, processes of standardization or variable adjustment used during analysis, and the calculation of effect sizes that we could extract from publications. This common limitation of meta-analyses was addressed by calculating an SMD, in this case Hedges’ g, using the information available in each publication. Care was taken to be consistent in data extraction, which was crosschecked by 2 persons, as was SMD calculation. For this analysis, we used commonly measured outcomes and most frequently methodological analytic approaches across trials. Interpretation of SMD effect sizes may differ across trials because of differences in tools used and population assessed but the strength of this study approach is to estimate an overall intervention effect interpretable in SD units for any given population.
In a few cases, study level estimates seemed to differ in magnitude from most of the other trials. A limitation of the current analysis is that the relatively small number of studies made it difficult to interpret whether these studies were truly different or simply differed because of chance. However, we were able to explore heterogeneity by factors expected to influence intervention effects when groups of studies could be classified together for effect modification testing.
There is a need for additional evaluations of group delivery, and further research on how intervention benefits are modified by the implementation variables evaluated here as well as others that could not be included. This will require that future studies include consistent reporting of implementation methods and adaptations.40 There is also a need for follow-up studies to examine how differences in implementation methods affect sustainability of benefits.
In conclusion, the Reach Up program, adapted across 8 countries, delivered by home visits or small groups and at varying scale, had benefits for child development and home stimulation in the majority of trials. Targeting undernourished children yielded larger benefits and beginning interventions when children are 12 months or older did not diminish impact. Findings suggest small group delivery is a promising strategy for scaling and that interventions of 12 months’ duration yield benefits as good or greater than longer programs. This evidence on implementation methods that affect impact can inform decision making as programs are taken to scale.
Acknowledgments
We thank study authors for assisting with requests for additional information including information for the subgroup analysis.
Dr Jervis conducted the meta-analysis, contributed to interpretation of findings, writing the analysis and results section, and critically reviewed the manuscript for crucial intellectual content; Dr Coore-Hall conducted the literature search and screening of studies for inclusion, conducted data extraction, assessment of risk of bias and quality of evidence, drafted the methods section, and contributed to critical review of the manuscript for crucial intellectual content; Dr Pitchik conducted data extraction, assessment of risk of bias and quality of evidence, drafted the methods section, and contributed to critical review of the manuscript for crucial intellectual content; Dr Arnold led the analysis and contributed to the interpretation of findings and writing of the analysis section and discussion; Dr Grantham-McGregor contributed to the conceptualization and design of the review, interpretation of findings, writing of the introduction and discussion, and critical review of manuscript for crucial intellectual content; Dr Marta Rubio-Codina contributed to design of the review, interpretation of findings, writing of the discussion, and critical review of the manuscript for crucial intellectual content; Drs Baker-Henningham, Fernald, Hamadani, and Trias contributed to design of the review and interpretation of findings and critical review of the manuscript for crucial intellectual content; Dr Smith conducted the literature search and screening of studies for inclusion and contributed to design of the review and critical review of the manuscript for crucial intellectual content; Dr Walker led the review, contributed to conceptualization and design, interpretation of findings, writing of introduction and discussion, and critical review of manuscript for crucial intellectual content; and all authors approved the final manuscript as submitted and agree to be accountable for all aspects of the work.
FUNDING: The study received support from the LEGO Foundation through a grant to the University of the West Indies. The Foundation had no role in the conduct of the review. Dr Jervis gratefully acknowledges financial support from the Institute for Research in Market Imperfections and Public Policy MIPP (ICS13_002 ANID) and the Center for Research in Inclusive Education, Chile (SCIA ANID CIE160009). The views presented do not represent the Inter-American Development Bank, its board of directors, or the countries they represent.
CONFLICT OF INTEREST DISCLOSURES: Dr Sally Grantham-McGregor developed the original Jamaica stimulation intervention. Drs Sally Grantham-McGregor, Susan P Walker, Jena Hamadani, Helen Baker-Henningham, Marta Rubio-Codina, and Joanne A Smith were involved in the development of the intervention as Reach Up. All authors have no financial relationships relevant to this article to disclose.
Comments