Parent-focused behavioural interventions for the prevention of early childhood obesity (TOPCHILD): a systematic review and individual participant data meta-analysis

Summary

Background

Childhood obesity is a global public health issue, which has prompted governments to invest in prevention programmes. We aimed to investigate the effectiveness of parent-focused early childhood obesity prevention interventions globally.

Methods

We did a systematic review and individual participant data meta-analysis. We searched databases and trial registries (MEDLINE, Embase, CENTRAL, CINAHL, PsycInfo, ClinicalTrials.gov, and WHO International Clinical Trials Registry Platform) from inception until Sept 30, 2024, for randomised controlled trials commencing before 12 months of age examining parent-focused behavioural interventions to prevent obesity in children, compared with usual care, no intervention, or attention control. Individual participant data were checked, harmonised, and assessed for integrity and risk of bias. We excluded trials that were quasi-randomised, investigated pregnancy-only interventions, or did not collect any child weight-related outcomes. The primary outcome was BMI Z score at age 24 months (±6 months). We did an intention-to-treat, two-stage, random effects meta-analysis to examine effects overall and for prespecified subgroups. We assessed certainty of evidence using Grading of Recommendations Assessment, Development, and Evaluation. This study is registered with PROSPERO, CRD42020177408.

Findings

Of 19 990 identified records, 47 (0·24%) trials were completed and eligible. Of these, 18 (38%) assessed our primary outcome, BMI Z score. We obtained individual participant data for 17 (94%; n=9128) of these 18 trials (n=9383), representing 97% of eligible participants. Of these 9128 participants, 4549 (50%) were boys, 4415 (48%) were girls, and 164 (2%) had unknown sex. We found no evidence of an effect of interventions on BMI Z score at age 24 months (±6 months; mean difference −0·01 [95% CI −0·08 to 0·05]; high certainty evidence, τ²=0·01; n=6505; 2623 missing). Findings were robust to prespecified sensitivity analyses (eg, different analysis methods and missing data), and we found no evidence of differential intervention effects for prespecified subgroups including priority populations and trial-level factors.

Interpretation

These findings indicate that examined parent-focused behavioural interventions are insufficient to prevent obesity at age 24 months (±6 months). This evidence highlights a need to re-think childhood obesity prevention approaches.

Funding

Australian National Health and Medical Research Council.

Introduction

The UN’s Sustainable Development Goals identify prevention of non-communicable diseases as a core priority.¹ Overweight and obesity in childhood are risk factors for non-communicable disease across the life course,^2,3 and represent major public health challenges that threaten recent health improvements contributing to increased life expectancy in many countries.⁴ Globally, around 37 million children younger than 5 years already live with overweight or obesity.² High BMI in infancy and very early childhood strongly predicts lifelong overweight and obesity.^4–6 Obesity in childhood tracks into adulthood,^7,8 and is linked to increased risk of short-term and long-term negative health consequences.^9,10 This large burden on health and social care systems has significant economic consequences for individuals and society.^11,12 Obesity prevalence rates are increasing disproportionately among socially disadvantaged populations.^2,13

To prevent obesity, many argue it is imperative to intervene early, when biology is most amenable to change, before overweight or obesity first develop in early childhood and obesity-conducive behaviour patterns are established.⁴ Consequently, the WHO report of the Commission on Ending Childhood Obesity⁴ recommends a life-course approach to reduce the risk of obesity, starting during preconception and pregnancy, and continuing throughout infancy and early childhood up to adolescence. WHO also highlights the need to “provide guidance on, and support for, healthy diet, sleep and physical activity in early childhood”⁴ for parents and caregivers, making early parent-focused interventions a key strategy for obesity prevention.

Previous evidence syntheses of early interventions for the prevention of obesity targeting child dietary, activity, and sleep behaviours have yielded mixed and inconclusive results. A systematic review¹⁴ assessing behavioural, educational, or quality improvement-based obesity prevention interventions in children younger than 2 years found no intervention improved child bodyweight status. The latest Cochrane review to examine the broader age group of children aged 0–5 years¹⁵ found that combined diet and physical activity interventions led to a small reduction in BMI Z score (mean difference −0·07 [95% CI −0·14 to −0·01]; 16 trials, n=6261, τ²=0·01, random effects, moderate certainty). A more recent update for the 2–4-year age group found that combined diet and physical activity interventions might result in a slight reduction in BMI Z score at long-term follow-up (>15 months; mean difference −0·07 [95% CI −0·13 to −0·01; ten trials, n=4693, τ²=0·00, random effects, low certainty).¹⁶ To put this into perspective, an effect of at least 0·2 BMI Z score units is generally considered clinically meaningful.¹⁷ Across these reviews, heterogeneity of interventions and reliance on published data limited the ability to conduct meaningful comparisons and sufficiently powered subgroup analyses. An individual participant data prospective meta-analysis of four Australian and New Zealand trials (commencing between 2007 and 2009) sought to address this gap, finding a small effect on BMI Z score at age 2 years (adjusted mean difference −0·12 [95% CI −0·22 to −0·02]; four trials, n=2196, heterogeneity p=0·09, fixed effects),¹⁸ although this effect dissipated at longer-term follow-up and when accounting for missing data.¹⁹

We have since identified a plethora of completed, ongoing, or planned randomised trials examining the effect of parent-focused interventions commencing before 12 months of age for primary prevention of childhood obesity. Such interventions are highly varied in their content, timing, dose, and populations reached.^20,21 Additionally, many of these programmes focused on bodyweight-related behavioural outcomes and were thus underpowered to detect a difference in anthropometric outcomes such as BMI. Uncertainty around the effectiveness of such interventions has put policy makers in a challenging position, resulting in decisions to implement potentially resource-intensive programmes that are based on insufficient evidence.^22–24

Individual participant data meta-analyses are the gold standard for combining randomised trial data.²⁵ Such studies enable harmonisation of outcome measures that are otherwise too heterogenous to be synthesised in aggregate form, increasing statistical power (eg, calculating BMI Z scores using a consistent reference population). They also allow examination of subgroup effects—ie, effect modification of intervention effects for different populations (eg, whether interventions work better for families with low socioeconomic status), with improved power and reduced risk of aggregation bias,²⁶ enabling us to understand what works for whom.²⁷ This feature is important, since childhood obesity prevention is a health equity issue,⁵ whereby families experiencing socioeconomic disadvantage are disproportionately affected by early childhood obesity and face greater barriers to accessing preventive interventions.^20,28 For instance, children living in the most socioeconomically disadvantaged areas of England are twice as likely to start school with obesity and three times as likely to start school with severe obesity compared with those from higher socioeconomic backgrounds.²⁹ We did a systematic review and individual participant data meta-analysis to assess whether parent-focused interventions commencing antenatally or in the first year after birth are effective in improving BMI Z score and behavioural outcomes at 2 years of age, and whether effectiveness varies across subgroups defined by individual-level or trial-level characteristics.

Methods

Detailed methods were prespecified in a published protocol,³⁰ a PROSPERO registration record (CRD42020177408), and a statistical analysis plan that was uploaded to Open Science Framework a priori (appendix pp 8–119). Minor protocol deviations with justifications (eg, limited data availability) are described in the appendix (pp 20–21). We followed PRISMA-IPD and PRISMA 2020 reporting guidelines (appendix pp 120–127).^31,32 Consumer representatives supported protocol development and interpretation of results (appendix p 7). The study protocol was approved by The University of Sydney Human Research Ethics Committee (2020/273).

Search strategy and selection criteria

We included individual and cluster-randomised controlled trials comparing parent-focused behavioural interventions for early childhood obesity prevention with usual care, no intervention, or attention control, that assessed at least one bodyweight-related outcome. We classified attention controls as those designed to match the intervention for contact time but with content not intended to affect key study outcomes (eg, general home safety information). We excluded quasi-randomised studies and those with active controls such as enhanced usual care or alternative interventions (ie, those designed to specifically affect the same outcomes as the experimental intervention) as our aim was to assess overall (rather than comparative) effectiveness and including these trials could dilute effect estimates. Eligible participants were parents or caregivers and their infants. Interventions could commence before or after birth, but needed to include intervention exposure between birth and 12 months of age. For trials including children both younger and older than 12 months at baseline, only data from children younger than 12 months were included. Eligible interventions focused on primary prevention of obesity in children and targeted at least one modifiable child behaviour related to obesity risk (ie, early feeding, diet, physical activity, and sleep). There were no language restrictions.

We systematically searched databases (MEDLINE, Embase, CENTRAL, CINAHL, and PsycInfo) from inception to Feb 27, 2023, and clinical trial registries (ClinicalTrials.gov and WHO International Clinical Trials Registry Platform) from inception to March 28, 2023 (both searches updated Sept 30, 2024), according to recommended guidelines³³ (appendix pp 128–136). We also consulted collaborators and hand-searched conference abstracts and reference lists of relevant reviews.

Two reviewers independently screened each study, with conflicts resolved by a third (KEH, SL, JGW, MA, JA, BJJ, AB, NS, or ALS). Trialists were consulted if uncertainties remained. Principal Investigators of all eligible trials were invited to join the Transforming Obesity Prevention for Children (TOPCHILD) Collaboration, give feedback on the protocol and statistical analysis plan, share their individual participant data, and contribute to a complementary intervention coding project.^21,34

Data collection, management, and analysis

Definitions of all outcomes, covariates, and subgroups were prespecified in the statistical analysis plan (appendix pp 41–67). Our primary outcome was BMI Z score at age 24 months (or within 6 months either side), using WHO Child Growth Standards.³⁵ We chose this continuous measure as our primary outcome because it is a strong predictor of lifelong obesity,^4–6 is commonly used in this age group,^15,18 and is considered the most feasible option for large-scale, population-level monitoring of obesity.^36,37 We prespecified several subgroups (appendix pp 56–62) at the individual level (eg, maternal education status, household income, and parity) and at the intervention level and trial level (eg, mode of intervention delivery, duration, and setting) to assess differential treatment effects for our primary outcome.²⁹

Prespecified key secondary outcomes (appendix pp 39–41) were duration of exclusive breastfeeding up to 6 months (or within 2 months either side), and each of the following at age 24 months (or within 6 months either side): daily vegetable consumption, screen time, physical activity, sleep duration, and parent feeding practice domain of control (restriction). Other prespecified secondary outcomes included other measures related to bodyweight status, feeding, dietary intake, activity, sleep, parent or carer behaviours, and adverse events (full list of secondary outcomes in appendix pp 39–41). To harmonise variables and outcomes, trial-level data were mapped, compared, and discussed at domain-specific outcome harmonisation workshops involving content and methods experts (see appendix pp 39–40, 115–117, 237–239 for further details and an illustrative example).

De-identified individual participant data were supplied by trial representatives for published and unpublished studies according to a prespecified data collection form (appendix pp 225–230). To ensure high-quality datasets, we conducted rigorous prespecified processing and cleaning procedures (appendix pp 22–25).

Using trial publications, registration records, and individual participant data, two reviewers independently assessed trial integrity with the IPD Integrity Tool,³⁸ conducted data quality checks according to a prespecified checklist derived from a scoping review,³⁹ and assessed risk of bias for the primary outcome and secondary outcome categories using the Cochrane Risk of Bias 2 Tool (appendix pp 137–43).^40,41 Discrepancies were resolved by a third reviewer, or by consulting trialists or advisors.

Once finalised, individual participant data from all trials were merged into a single dataset. If individual participant data were not obtained for a trial, aggregate data were extracted from published records by two reviewers, with conflicts adjudicated by a third. We followed a prespecified checklist⁴² to determine whether to include published aggregate data when individual participant data were unavailable (appendix p 144).

Certainty of evidence was assessed by two reviewers using the Grading of Recommendations Assessment, Development and Evaluation approach⁴³ for the primary outcome and key secondary outcomes.

All analyses were prespecified in a time-stamped statistical analysis plan (appendix pp 8–119) and done using R (version 4.4.1) or Stata (release 18) for floating subgroup analyses (metafloat package).²⁶ All analyses were intention to treat following the principles of White and colleagues,⁴⁴ including all eligible randomly assigned participants for whom data were available.

Analysis for all outcomes was conducted using a complete-case two-stage approach, assuming random effects in generalised linear mixed models. In accordance with the decision-tree prespecified in our analysis plan (appendix p 71) and best-practice guidance,⁴⁵ we favoured a two-stage over a one-stage approach to overcome concerns around model stability, convergence issues, and to minimise the risk of aggregation bias. For cluster-randomised trials, correlated data were accounted for by adding random intercepts within trials. We adjusted for sex, as a fixed (common) effect. For continuous outcomes, we estimated mean differences. For binary outcomes and count outcomes, we estimated risk ratios or rate ratios (ie, log link) or, if binary outcome models failed to converge, we estimated odds ratios (ie, logit link). For binary outcomes, when no events were recorded in a trial for both the control and intervention groups, that trial was not included for that outcome in line with best-practice recommendations.⁴⁶ Trials that contained no events in one group and events in the other, so-called single-zero trials, were also removed because of convergence issues (but assessed in post-hoc sensitivity analyses; appendix pp 205–07). For breastfeeding duration outcomes, we estimated hazard ratios using Cox regression models to account for infants still breastfeeding at study end. To account for multiplicity issues, outcomes were interpreted as patterns of evidence in consideration of clinical plausibility, rather than focusing on any single statistically significant result.⁴⁷ Our complete statistical model and example code are available in the appendix (pp 234–36).

Heterogeneity was examined by inspecting forest plots, τ² (ie, the estimated variance of true effect sizes across studies in a random-effects meta-analysis), and 95% prediction intervals (PIs; the range within which the treatment effect in a future trial from a similar population is expected to fall), and by conducting a common effects sensitivity analysis. We did individual-level subgroup analyses (ie, effect modification analyses) for the primary outcome by examining within-trial treatment-by-covariate interactions with common effects to avoid aggregation bias.⁴⁸ We did trial-level subgroup analyses for continuous variables using two-stage random effects meta-regressions, with a single moderator. We compared trial-level subgroup analyses for categorical variables using a Q_m test to assess whether the moderator coefficients significantly differed from the null. Prespecified sensitivity analyses for the primary outcome included different analysis methods (one-stage model with a random treatment effect stratified by trial and centre, two-stage fixed effects model, incorporation of aggregate data where individual participant data were unavailable, exclusion of trials with high risk of bias, integrity concerns, significant conflict of interest, >40% missing data, or low adherence [trial-level], and multiple imputation; appendix pp 72–74). We did multiple imputation using the mice package in R (appendix p 72).⁴⁹

Role of the funding source

The funder of this study had no role in study design, data collection, data analysis, data interpretation, or writing of the report.

Results

We identified 19 990 records from database and registry searches (figure 1). Of these, 8030 (40·2%) were duplicates, and 10 737 (53·7%) were excluded after title and abstract screening. For the remaining 1223 (6·1%) records, we reviewed full-text reports and identified 47 (3·8%) eligible and completed trials (appendix pp 145–79). Of these, 18 (38%) trials assessed the primary outcome for the meta-analysis of BMI Z score at age 24 months (or within 6 months either side) and thus were eligible for our primary outcome analysis. We obtained individual participant data for 17 (94%) of these 18 trials, representing 9128 (97%) of 9383 total eligible participants. Of these 9128 participants, 4549 (49·8%) were boys, 4415 (48·4%) were girls, and 164 (1·8%) had unknown sex. We did not include aggregate data for the one trial without individual participant data following prespecified decision rules (appendix pp 68–70, 144). 31 (91%) of the 47 trials contributed data to secondary outcomes (table 1).

Figure 1:

Study profile

Table 1:

Characteristics of included studies (n=31)

	Dates	Sample size*	Unit of randomisation	Participants	Intervention period	Intervention	Control	Primary outcome
Trials contributing to primary outcome analysis, with or without secondary outcome analysis (n=17)
Australia (Campbell et al 2013)50	2008–10	Randomly assigned: 542 Eligible: 542	Cluster	First-time parents regularly attending first-time parent group Parent groups were eligible if ≥8 parents were enrolled, or ≥6 parents were enrolled in areas of low socioeconomic position	Began within the first 6 months; ended by age 24 months	Six 2 h sessions delivered within pre-existing mothers’ groups (age 3, 6, 9, 12, 15, and 18 months) Behaviours targeted: food provision and parent feeding practices and movement practices (n=271)	Usual care plus quarterly newsletter on general child health messages excluding sleep, food, and activity (n=271)	Dietary intake
Australia (Campbell et al 2016)51	2011–20†	Randomly assigned: 514 Eligible: 514	Cluster	First-time parents with infants aged 3–4 months regularly attending a parent group in disadvantaged areas	Began within the first 6 months; ended by age 36 months	Six 1·5 h group sessions (age 3, 6, 9, 12, 15, and 18 months), then six quarterly newsletters (21, 24, 27, 30, 33, and 36 months) Behaviours targeted: food provision and parent feeding practices and movement practices (n=263)	Usual care plus quarterly generic child health newsletters (n=251)	Anthropometry: height, bodyweight, waist circumference, and BMI Z score at age 18 months and 36 months
Australia (Daniels et al 2013)52	2008–10	Randomly assigned: 698 Eligible: 698	Individual	First-time mothers of healthy term infants	Began at age 4–7 months; ended by age 16 months	Two education peer support modules (six fortnightly sessions each) at age 4–7 months and 13–16 months at community health venues Behaviours targeted: food provision and parent feeding practices and movement practices (n=346)	Usual care plus quarterly newsletter on general child health messages excluding sleep, food, and activity (n=352)	Food intake, food preferences, and feeding behaviour
Australia (Wen et al 2012)53	2007–10	Randomly assigned: 667 Eligible: 667	Individual	Women aged 16 or over, expecting their first child, in 24–34 weeks’ pregnancy	Began during pregnancy; ended by age 24 months	Eight home visits providing maternal advice (antenatal and at age 1, 3, 5, 9, 12, 18, and 24 months) Behaviours targeted: infant feeding practices, food provision and parent feeding practices, and movement practices (n=337)	Usual care plus written home safety and tobacco prevention information sent by post at 6 months and 12 months (n=330)	BMI at age 2 years
Australia (Wen et al 2022)54	2017–20	Randomly assigned: 1155 Eligible: 1155	Individual	Women aged ≥16 years in their third trimester that can communicate in English, Chinese, or Arabic	Began during pregnancy; ended by age 24 months	Group 1 (telephone support): nine staged intervention booklets (posted) and nine 30–60 min telephone support sessions to mothers by Child & Family Health Nurses (third trimester, and at age 1, 3, 5, 7, 10, 12–15, 15–18, and 18–24 months; n=386) Group 2 (SMS): nine staged SMS interventions after posting intervention booklets Behaviours targeted: infant feeding practices, food provision and parent feeding practices, and movement practices (n=384)	Usual care comprising at least one nurse visit for general support at home and possible multiple home visits for vulnerable families from the local health districts (n=385)	BMI, breastfeeding duration, and timing of introduction of solids
Italy (Morandi et al 2019)55	2014–17	Randomly assigned: 529 Eligible: 529	Cluster (paediatrician)	Healthy term newborn babies	Began in the first 6 months; ended by age 24 months	Intensive education: at each well visit until child age 2 years, parents provided with oral and written information on obesity-protective behaviours for their children Behaviours targeted: infant feeding practices, food provision and parent feeding practices, and movement practices (n=278)	Usual education about nutrition and lifestyle, during their child’s first 2 years of life (n=251)	BMI at age 2 years
Netherlands (Karssen et al 2021)56	20l8–19†	Randomly assigned: 357 Eligible: 270	Individual	Parents with low or medium educational attainment with an infant aged 5–15 months	Began at age 5–15 months; ended after 12 months	Mobile application parenting programme, Samen Happie!, teaching parents about healthy parenting practices and general healthy authoritarian parenting style Behaviours targeted: food provision and parent feeding practices, movement practices, and sleep health practices (n=137)	Waiting list control (n=133)	BMI at 6 months and 12 months post baseline
New Zealand (Taylor et al 2017a)57	2009–17	Randomly assigned: 802 Eligible: 802	Individual	Women aged >16 years before 34 weeks’ gestation; preterm infants born before 36·5 weeks excluded	Began during pregnancy; ended by age 24 months	Group 1 (food, activity and breastfeeding): mix of seven home visits and group-based sessions promoting breastfeeding, healthy eating, and physical activity (1 week, and age 3, 4, 7, 9, 12, and 18 months; n=214) Group 2 (sleep): two home visits (antenatal and 3 weeks) targeting prevention of sleep problems, as well as a sleep treatment programme if requested (age 6–24 months; n=209) Group 3: (food, activity and breastfeeding, and sleep) Behaviours targeted: infant feeding practices, food provision and parent feeding practices, movement practices, and sleep health practices (n=210)	Usual care: first 4-week midwife home visits Well Child (Plunkett) nurse: eight visits in 5 years (n=214)	Bodyweight at age 6, 12, and 24 months and BMI at age 24 months
New Zealand (Taylor et al 2017b)58	2012–16	Randomly assigned: 206 Eligible: 206	Individual	Pregnant women aged ≥16 years, before 34 weeks’ gestation, who could communicate in English or Te Reo Maori; preterm infants excluded	Began during pregnancy; ended by age 12 months	Baby-led introduction to solids: lactation consultant support (three face-to-face sessions and two telephone sessions; 10–60 min each) up to age 6 months and three personalised face-to-face contacts (age 5·5, 7, and 9 months) Behaviours targeted: infant feeding practices, food provision, and parent feeding practices (n=105)	Usual Well Child care, usually 6–7 home or clinic visits (from age 6 weeks to 2 years) from a trained health professional (n=101)	BMI at age 12 months
Norway (Helle et al 2019)59	2015–21	Randomly assigned: 533 Eligible: 533	Individual	Parents and their child aged 3–5 months	Began at age 6 months; ended by age 12 months	eHealth intervention: access to website with seven monthly short video clips (3–5 min) addressing infant feeding topics and age-appropriate baby food recipes Behaviours targeted: food provision and parent feeding practices (n=269)	Usual care from the local child health clinic with consultations at child age 6, 8, 10, and 12 months (n=264)	Child eating behaviour, food intake, mealtime routines, and maternal feeding practices
Norway (Øverby et al 2017)60	2012–15	Randomly assigned: 110 Eligible: 110	Individual	Parents and their 4–6-month-old infants attending selected public health clinics	Began in the first 6 months; ended by age 6 months	Two 4-h course days providing parent groups with nutritional information and instruction to prepare nutritious and varied dishes, delivered by home economics teacher and Masters student Behaviours targeted: food provision and parent feeding practices (n=56)	Parents received a booklet containing recipes for homemade foods for infants (n=54)	Food intake at age 6, 15, and 24 months
Norway (Røed et al 2021)61	2015–22	Randomly assigned: 298 Eligible: 237	Individual	Infants close to age 12 months and one of their parents	Began in the first 12 months; ended by age 24 months	Food4toddlers eHealth intervention: website with seven modules (2–4 lessons approximately 10 min each) promoting healthy food and eating environments, recipes, discussion forum, information about food and beverages, plus 20 weekly emails with link to new lessons Behaviours targeted: food provision and parent feeding practices (n=148)	Usual care at the community child health centres (n=150)	Child diet quality and food variety, assessed at age 18 months and 24 months
Sweden (Döring et al 2016)62	2008–15	Randomly assigned: 1369 Eligible: 1148	Cluster (child health-care centres)	First-time mothers and their children (aged 9–10 months) recruited at child health-care centres	Began in the first 12 months; ended by age 48 months	Nine sessions: one group (11 months), six individual (age 8–9 months, and 1, 1·5, 2, 3, and 4 years), two individual telephone sessions (age 2·5 years and 3·5 years) delivered by nurse focusing on healthy food habits and physical activity Behaviours targeted: food provision and parent feeding practices and movement practices (n=601)	Usual care: regular age-related health checkups of Swedish child health services (n=768)	BMI and waist circumference of children at age 4 years and their mothers
UK (Bryant et al 2021)63	2017–19	Randomly assigned: 117 Eligible: 28	Cluster (children’s centres)	Parents or other carers and at least one child aged 6 months to 5 years	Began within the first 6 months to 5 years; ended after 8 weeks	8-week Health, Exercise, Nutrition for the Really Young programme, including eight weekly 2·5 h sessions delivered in children centres to groups of 8–10 parents Behaviours targeted: food provision and parent feeding practices and movement practices (n=47)	Waiting list control (n=70)	Feasibility and child BMI Z score
USA (Messito et al 2020)64	2012–20	Randomly assigned: 533 Eligible: 533	Individual	Latina mothers with a singleton uncomplicated pregnancy, fluent in English or Spanish	Began during pregnancy; ended by age 36 months	Starting Early Program: 15 sessions, two individual (third trimester and postpartum) and 13 group (age 1, 2, 4, 6, 9, 12, 15, 18, 21, 24, 27, 30, and 33 months), providing nutrition counselling and support Behaviours targeted: infant feeding practices and movement practices (n=266)	Usual care: one prenatal nutrition consultation, one childbirth or breastfeeding class, as-needed lactation support, and paediatric visits as per American Academy of Pediatrics guidelines (n=267)	Infant feeding practices and material infant feeding knowledge
USA (Paul et al 2018)65	2012–23	Randomly assigned: 291 Eligible: 291	Individual	Full-term singleton infants born to primiparous mothers	Began within the first 6 months; ended by age 36 months	Responsive parenting: four home visits by research nurses (age 3, 16, 28, and 40 weeks) and annual research centre visits until age 3 years, focused on feeding, sleep, interactive play, and emotion regulation Behaviours targeted: infant feeding practices, food provision and parent feeding practices, movement practices, and sleep health practices (n=145)	Home safety intervention (n=146)	BMI Z score at age 3 years
USA (Sanders et al 2021)66	2010–14	Randomly assigned: 865 Eligible: 865	Cluster (trial sites)	Infants presenting for 2-month well-child check-up, whose caregiver could speak Spanish or English	Began at age 2 months; ended by age 18 months	Greenlight toolkit: low literacy booklets that reviewed dietary, physical activity, sleep, and screentime advice for parents and education for providers on health communication Behaviours targeted: infant feeding practices, food provision and parent feeding practices, movement practices, and sleep health practices (n=459)	Attention placebo: injury prevention counselling according to The injury Prevention Program by the American Academy of Pediatrics (n=406)	Percentage of children with overweight or obesity at age 2 years
Other trials contributing to secondary outcomes only (n=14)
Belarus (Kramer et al 2001)67	1996–98	Randomly assigned: 17 046 Eligible: 17 046	Cluster (hospital and associated outpatient clinic)	Full-term singleton infants weighing at least 2500 g and their healthy mothers who intended to breastfeed	Began at birth; ended by age 12 months	Breastfeeding promotion and support according to the WHO’s Baby Friendly Hospital Initiative at hospital and follow-up visits Behaviours targeted: infant feeding practices (n=8865)	Usual care (n=8181)	Breastfeeding (any and duration), gastrointestinal infection, respiratory tract infection, and atopic eczema during the first 12 months of life
Brazil (Sangalli et al 2021)68	2008–10	Randomly assigned: 715 Eligible: 715	Cluster (health-care centre)	Pregnant women in their third trimester attending health centres predominantly serving low-income families	Began during pregnancy; ended by age 24 months	Breastfeeding promotion, introduction of foods, healthy eating and healthy eating habits, based on the Ten Steps for Healthy Feeding guideline Behaviours targeted: infant feeding practices, food provision and parent feeding practices (n=373)	Usual care (n=363)	Exclusive breastfeeding at age 4 months
UK (Lakshman et al 2018)69	2011–15	Randomly assigned: 669 Eligible: 669	Individual	Parents (mainly mothers) and their infants (aged 2–14 weeks) who were formula-fed	Began within the first 6 months; ended by age 6 months	Baby Milk programme supported mothers feeding their babies according to the WHO recommendations for energy requirements: three 30–45 min face-to-face contacts (at age 2, 4, and 6 months) and two 15–20-min telephone calls (at age 3 months and 5 months) and leaflets (at age 2 months and 4 months); involved components on motivation, setting goals and actions, and overcoming barriers Behaviours targeted: infant feeding practices (n=340)	Usual care group had the same number of contacts but received general information about formula-milk feeding and infant health (n=329)	The change in infant bodyweight SD score from birth to age 12 months
UK (McEachan et al 2016)70	2012–14	Randomly assigned: 120 Eligible: 120	Individual	Pregnant women with overweight or obesity (BMI ≥25 kg/m2) at 10–12 weeks’ gestation and infants from birth	Began during pregnancy; ended by age 12 months	Healthy and Active Parenting Programme for Early Years aimed to promote breastfeeding, promote healthy eating and habits, and promote physical activity: delivered through 12 group sessions (six antenatal and six postnatal); the intervention was developed to be culturally appropriate for key groups (White British and South Asian origin women) Behaviours targeted: infant feeding practices, food provision and parent feeding practices, and movement practices (n=59)	Usual care (n=6l)	Child bodyweight at age 12 months
USA (de la Haye et al 2019)71	2018–19	Randomly assigned: 50 Eligible: 50	Individual	Mother–child dyads enrolled in home visitation programmes—included mothers facing poverty, housing instability, lack of social and material support, lack of transportation, and limited education and literacy	Began after birth; ended after 6 months	Home visitation programme core curriculum with nutrition and physical activity enhancement: home visits (weekly for 6 months) Behaviours targeted: food provision and parent feeding practices and movement practices (n=30)	Healthy families America Home visitation programme: home visits (weekly, up to age 2–5 years)—a culturally sensitive programme to strengthen parent-child relationships, promote child development, and link to community resources (n=20)	Bodyweight of mothers, rate of weight gain of infants, and waist circumference of mother
USA (Fiks et al 2017)72	2014–15	Randomly assigned: 87 Eligible: 85	Individual	Pregnant women with overweight or obesity (BMI ≥25 kg/m2), Medicaid insured, and who owned a smartphone	Began during pregnancy; ended by age 12 months	Grow2Gether for healthy infant growth and behaviour: two in-person meetings (at enrolment and age 4 months) and 11 online group activities (two prenatally, and until age 9 months) Behaviours targeted: infant feeding practices, food provision and parent feeding practices, movement practices, and sleep health practices (n=43)	Parents received text message reminders to schedule recommended primary care visits for their infant, and to attend appointments scheduled in Children’s Hospital of Philadelphia Care Network (n=42)	Feasibility
USA (Linares et al 2019)73	2016–18	Randomly assigned: 39 Eligible: 39	Individual	Immigrant Hispanic pregnant women who intended to breastfeed their infants	Began during pregnancy; ended by age 6 months	Early Childhood Obesity Risk-Reduction Program in Hispanics culturally acceptable and linguistically diverse promotion of exclusive breastfeeding: prenatal sessions (40 min), prenatal calls (10 min), hospital visit (30 min), postpartum home visit (40 min), and postpartum calls (monthly, 10 min) Behaviours targeted: feeding practices (n=20)	Usual care provided by the Special Supplemental Nutritional for Women, Infants, and Children programme (n=19)	Exclusive breastfeeding from birth to age 6 months
USA (Palacios et al 2018)74	2016–16	Randomly assigned: 202 Eligible: 202	Block	Caregivers of healthy term infants aged 0–2 months participating in the WIC Program in Puerto Rico and Hawaii	Began within the first 2 months; ended after 4 months	Weekly text messages (for 4 months) reinforcing the feeding messages provided by WIC Behaviours targeted: infant feeding practices and food provision and parent feeding practices (n=102)	Control text messages were sent, relating to general infant’s health (n=100)	Infant weight-for-length percentile
USA (Paul et al 2011)75	2006–09	Randomly assigned: 160 Eligible: 160	Individual	Mother-newborn baby dyads, primiparous, singleton, and gestational age ≥34 weeks	Began 2–3 weeks after birth; ended by age 6 months	SLIMTIME Nurse home visits (2–3 weeks after birth and at age 4–6 months) Group 1 (introduction to solids): instruction on delay of complementary foods and importance of repeat exposure to foods Behaviours targeted: infant feeding practices and food provision and parent feeding practices (n=38) Group 2 (soothe/sleep): parents were taught alternate strategies to feeding as an indiscriminate first response to infant distress Behaviours targeted: infant feeding practices and sleep health practices (n=39) Group 3 (soothe/sleep and introduction to solids): both Group 1 and Group 2 interventions Behaviours targeted: infant feeding practices, food provision and parent feeding practices, and sleep health practices (n=42)	Usual care (n=41)	Weight-for-length percentile at age 1 year
USA (Rybak et al 2023)76	2021–22	Randomly assigned: 65 Eligible: 65	Individual	Mothers of singleton infants born >2500 g, at 37–42 weeks’ gestation, English-speaking, and attending a primary care practice primarily serving minority groups and low-income populations	Began at 1 month well-child visit; ended at 6-month well-child visit	A strengths-based responsive parenting intervention, teaching healthy responsive parenting during infancy to promote vital growth and regulation, delivered via Integrated Behavioral Health at well-child visits (at age 1, 2, 4, and 6 months) that helps caregivers recognise and respond to infant cues for hunger, fullness, and distress, emphasising non-feeding soothing strategies, responsive feeding practices, and sleep-promoting behaviours Behaviours targeted: infant feeding practices, food provision and parent feeding practices, and sleep health practices=33)	A socioemotional development and positive parenting intervention matched for time, attention, and level of provider, but without the specific active ingredients of the THRIVE intervention related to infant feeding, sleep, and regulation (n=32)	Feasibility (enrolment) and acceptability (retention and adherence)
USA (Stough et al 2018; NCT03597061)	2018–20†	Randomly assigned: 34 Eligible: 32	Individual	Parents and infants born at >38 weeks’ gestation, above 10th percentile of length-for-weight, and aged 2–3 months	Began at age 4 months; ended by age 9 months	Healthy Start to Feeding has three individual sessions providing parent education and skills training on responsive feeding approach to introduction of healthy foods Behaviours targeted: infant feeding practices and food provision and parent feeding practices (n=l6)	Participants and their parents completed pre-treatment and posttreatment period study visits to assess study outcomes; they received no intervention (n=16)	Weight-for-length percentile, appetite regulation, and fruit and vegetable variety at age 3 months and 9 months
USA (Thomson et al 2018)77	2013–16	Randomly assigned: 82 Eligible: 54	Individual	Pregnant women at least 18 years of age, <19 weeks pregnant and their infant from birth, residing in a rural region with high rates of infants with low birthweight, preterm infants, child poverty, and childhood overweight/obesity	Began during pregnancy; ended by age 12 months	Parents as Teachers Experimental group received the enhanced nutrition and physical activity lessons and materials, which follow the family wellbeing Parents as Teachers curriculum; the added maternal weight management and early childhood obesity prevention components are based on social cognitive theory and behaviour change; monthly lessons at in-home visits from gestational month 4 to age 12 months Behaviours targeted: infant feeding practices, food provision and parent feeding practices, movement practices, and sleep health practices (n=24)	Parents as Teachers curriculum, home visits, optional group sessions (monthly), developmental screenings, and resource network for families, which aimed to increase parental knowledge of child development, improve parenting, early detection of developmental delay, prevent abuse, and increase child reading (n=30)	Maternal outcomes: gestational bodyweight gain, bodyweight retention, dietary intake, and physical activity Infant outcomes: dietary at age 12 months
USA (Trak-Fellermeier et al 2019)78	2013–15	Randomly assigned: 31 Eligible: 31	Individual	Women with overweight/obesity, singleton pregnancy <16 weeks’ gestation and their infant from birth, residing in Puerto Rico	Began during pregnancy; ended 6 days after birth	Health empowerment programme: individual visits, two visits (prenatal at around 16 weeks’ and 27 weeks’ gestation); group sessions, six 2 h sessions (starting 1–2 weeks after randomisation, occurring every 2 weeks); and telephone calls, six 30 min calls (monthly) Behaviours targeted: infant feeding practices, food provision and parent feeding practices, movement practices, and sleep health practices (n=15)	Group sessions with health advice about dental care and child safety (n=16)	Gestational bodyweight gain
USA (Wasser et al 2020)79	2013–17	Randomly assigned: 430 Eligible: 429	Individual (stratified)	Primiparous and multiparous non-Hispanic Black mothers enrolled at 28 weeks’ pregnancy	Began during pregnancy; ended by age 12 months	Home visits: six visits by peer educators (prenatal <28 weeks’ gestation and antenatal at age 1, 3, 6, 9, and 12 months); and maternal and study partner advice Behaviours targeted: infant feeding practices, food provision and parent feeding practices, movement practices, and sleep health practices (n=215)	Attention-control received maternal advice on injury prevention (n=214)	Infants’ mean weight-for-length Z score at age 15 months

Allocation ratio was 1:1 for all trials. WIC=Women, Infants, and Children (a programme that supports high-risk, low-income populations).

^*Randomly assigned sample size represents the total number of participants randomly assigned in the trial. Eligible sample size represents the total number of participants randomised in the trial that met eligibility criteria for TOPCHILD (eg, Bryant and colleagues⁶³ randomly assigned 117 children aged 6 months to 5 years. Of these, 28 participants were randomly assigned before age 12 months and thus were eligible for TOPCHILD).

^†Main results unpublished at time of analysis.

The 17 trials in the primary outcome analysis were set across eight countries (all high-income).⁸⁰ Of these, five (29%) trials targeted families with low socioeconomic position, two (12%) targeted families from culturally and linguistically diverse backgrounds or ethnic minority groups and the remaining ten (59%) targeted a global population.²⁰ The median eligible sample size was 533 (IQR 270–698). Interventions in five (29%) of 17 trials commenced during pregnancy; postnatal interventions commenced at a median child age of 3·91 months (IQR 1·66–5·75). Mean maternal pre-pregnancy BMI was 25·33 (SD 5·45) and mean age at birth was 30·36 years (SD 5·34). Mean child birthweight was 3418·54 g (SD 529·83). Baseline characteristics were balanced, and missing data for the primary outcome (n=2623) were evenly distributed between intervention (28%) and control (29%) groups (appendix pp 180–82).

We found no evidence that the intervention had an effect on the primary outcome of BMI Z score at age 24 months (or within 6 months either side), with a mean difference of −0·01 (95% CI −0·08 to 0·05; p=0·15; high certainty of evidence; τ²=0·01 [95% PI −0·17 to 0·14]; n=6505; 2623 missing; table 2; figure 2). Results were consistent across all prespecified sensitivity analyses, including a complete case analysis excluding the four trials with more than 40% missingness (mean difference −0·02 [95% CI −0·09 to 0·04]; τ²=0·00; 13 trials; n=6170; 1808 missing), multiple imputation for missing outcome data (mean difference −0·02 [95% CI −0·08 to 0·04]; τ²=0·00; 17 trials), and excluding trials with high risk of bias (appendix pp 190–195, 207). Risk of bias was rated high for seven of 17 trials including 3246 (36%) participants (due to missing data), some concern for four trials including 2364 (26%) participants, and low for six trials including 3518 (39%) participants. We found no evidence that the effect on BMI Z score differed according to any of the 19 individual or trial-level or intervention-level subgroup characteristics (ie, no evidence of effect modification, including infant sex), except for a minor increase in intervention effect by 0·01 for every 100 g increase in birthweight (interaction mean difference 0·01 [95% CI 0·00 to 0·02]; tables 3–6). We found no evidence of publication bias from visual inspection of the funnel plot and Egger’s test (p=0·91; appendix p 189).

Table 2:

Primary and key secondary outcomes

	Trials	Participants	Intervention*	Control*†	Effect estimate (95% CI)	Heterogeneity, τ2 (95% PI)	GRADE
Primary outcome
BMI Z score at age 24 months (±6 months)	17	6505	0·71 (1·08)	0·72 (1·05)	Mean difference −0·01 (−0·08 to 0·05)	0·01 (−0·17 to 0·14)	High
Key secondary outcomes
Duration of exclusive breastfeeding assessed at age 6 months (±2 months), weeks	5	1653	12·22 (9·32); 13 (10 to 13)‡	10·69 (8·86); 12 (8·69 to 13)‡	Hazard ratio 0·86 (0·74 to 1·00)	0–01 (0·65 to 1·14)	Moderate
Vegetable intake per day at age 24 months (±6 months), g	12	4616	117·18 (90·06)	107·48 (84·09)	Mean difference 3·11 (−0·64 to 6·85)	0–00 (−0·64 to 6·85)	Moderate
Screen time per day at age 24 months (±6 months), min	9	3650	59·58 (69·98)	72·36 (69·98)	Mean difference −9·60 (−13·72 to −5·47)	0·00 (−13·72 to −5·47)	Moderate
Physical activity per day at age 24 months (±6 months), min	1	314	252·59 (137·46)	276·60 (150·13)	Mean difference −24·14 (−57·17 to 8·90)	NA	Very low
Combined sleep duration per night and day at age 24 months (±6 months), h	10	3839	12·68 (1·53)	12·69 (1·57)	Mean difference 0·06 (−0·02 to 0·15)	0·002 (−0·06 to 0·18)	Moderate
Parent feeding practices, control (restriction) score ≥3 at age 24 months (±6 months)§	2	545	81/269(30%)	91/276 (33%)	Risk ratio 0·90 (0·72 to 1·13)	0·00 (0·72 to 1·13)	Low

Data are n, n/N (%), mean (SD), or median (IQR), unless otherwise specified. GRADE=Grading of Recommendations Assessment, Development, and Evaluation. NA=not applicable. PI=prediction interval.

^*Means are crude estimates not adjusting for clustering by trial or centre.

^†For multi-arm trials, the approximate adjustment method was used to avoid unit-of-analysis errors that can be introduced when using the same control group for two different comparisons.⁸¹

^‡Survival analysis performed for this outcome, thus survival median with 95% CI is reported.

^§Event defined as regular use of feeding practice—ie, domain score of at least 3.⁸²

Figure 2: Forest plot for the primary outcome of BMI Z score at age 24 months (or within 6 months either side).

There are two multi-arm trials: Wen and colleagues⁵⁴ (2022) and Taylor and colleagues⁵⁷ (2017). Each group is indicated by A, B, or C after the study name. The black squares capture the intervention effect estimate. The 95% CIs around this estimate are represented by the black line. Lines with arrows indicate CIs that extend beyond the scale of the plot. The adjusted two-stage random effects model (bolded) presents the primary analysis. τ² is the variance of the between-study effects. The prediction interval represents the range within which the treatment effect in a future trial from a similar population is expected to fall. The adjusted two-stage fixed effects model and the adjusted one-stage random effects model are sensitivity analyses.

Table 3:

Subgroup analyses by individual-level characteristics for the primary outcome of BMI Z score at age 24 months (or within 6 months either side), continuous individual-level factors

	Studies	Participants	Pooled interaction effect (95% CI)
Birthweight, 100 g	16	6222	0·01 (0·00 to 0·02)
Gestational age at birth, weeks	7	3751	0·02 (−0·01 to 0·06)
Maternal BMI, kg/m2	15	6025	−0·00 (−0·02 to 0·01)
Weighted standardised household income*	6	2301	0·14 (−0·07 to 0·36)

Data are n, unless otherwise indicated. For continuous subgroups, the pooled interaction effect is the difference in intervention effect after a one-unit increase in the covariate, assuming a linear relationship between the intervention and BMI Z score across all levels of the covariate (eg, the intervention increases BMI Z score by 0.01 for every 100-g increase in birthweight). A result is statistically significant if the 95% CI of the pooled interaction effect does not include 0. For each participant, weighted standardised household income was calculated as the difference between total household income per year at baseline and annual median country and year specific household income, divided by annual median country and year specific household income (appendix p 60).

^*0 indicates earning the median, greater than 0 indicates earning more than the median, and below 0 indicates earning less than the median.

Table 6:

Subgroup analyses of trial-level characteristics for the primary outcome of BMI Z score at age 24 months (±6 months), continuous trial-level characteristics

	Studies	Participants	Meta-regression slope estimate (95% CI)
Intervention duration, h	15	6263	−0·010 (−0·023 to 0·004)
Development of country, human development index	17	6505	−0·70 (−6·43 to 5·04)
Intervention end, months	15	6239	0·0005 (−0·0056 to 0·0066)

Data are n, unless otherwise indicated.

For secondary outcomes, 47 (100%) completed trials were eligible, with 34 (72%) providing individual participant data. Three (9%) were excluded because of integrity issues relating to missing data and randomisation, leaving 31 (91%) trials contributing data to secondary outcomes (28 825 participants), although not all trials collected data on all outcomes (table 2). Mean maternal pre-pregnancy BMI was 25·62 kg/m² (SD 5·68) and mean age at birth was 26·67 years (SD 5·71). Mean child birthweight was 3422 g (SD 465). Baseline characteristics were balanced, with missing data evenly distributed across intervention and control groups (appendix pp 182–83).

For key secondary outcomes, we found no evidence that the intervention had an effect on duration of exclusive breastfeeding at age 6 months (or within 2 months either side), or on daily vegetable intake, physical activity, or sleep duration at age 24 months (or within 6 months either side; table 2). Small improvements were seen in screen time at age 24 months (or within 6 months either side), with an average decrease of 9 min and 36 s per day in the intervention group compared with control (nine trials; n=3650; mean difference −9·60 [95% CI −13·72 to −5·47]; moderate certainty).

For the remaining 27 secondary outcomes across the categories of anthropometry, feeding, diet, movement, sleep, parent practices, and adverse events, we found no differences between the intervention and control groups, except for a few isolated small effects (appendix pp 187–188). Risk of bias was rated high for most secondary outcomes (such as diet and physical activity) because of self-reporting and missing data. For anthropometric secondary outcomes, nine (31%) of 29 trials had low risk of bias due to objective, blinded measurements; six (21%) had some concerns; and 14 (48%) were rated high risk of bias due to missing data (appendix p 138).

Discussion

This large individual participant data meta-analysis of randomised controlled trials found no evidence of effectiveness of early childhood obesity prevention interventions for our primary outcome BMI Z score and most secondary outcomes around infant feeding, movement, diet, and sleep. There was a high certainty of evidence for the primary outcome, which included 9128 (97%) of 9383 eligible participants from 17 (94%) of 18 identified trials and was robust across subgroups and sensitivity analyses. Our findings align with previous reviews, which reported no effect¹⁴ or only very small effects below the threshold of clinical significance.^15,18 Despite the 17 trials being set in eight countries and targeting various populations, the findings were quite homogenous, with little heterogeneity across trials. Certainty of evidence varied for secondary outcomes and data were available from 31 (70%) of 44 identified trials (28 825 [86%] of 33 689 eligible randomly assigned participants).

This is the most comprehensive review of early obesity prevention interventions to date. We used gold-standard methods for meta-analysis, including rigorous checks of data quality, integrity, and risk of bias. Availability of individual participant data enabled harmonisation of outcomes with heterogeneous definitions and measurement methods (eg, portion sizes of fruits and vegetables across different countries; appendix pp 115–17), and inclusion of three unpublished studies and several unpublished outcomes, thereby increasing statistical power to provide the most reliable and precise effect estimates. The collection of individual participant data also enabled exploration of potentially differential effects across key individual and trial-level characteristics, addressing whether these interventions work exclusively or better in some populations than others.

Limitations should be considered. Seven (41%) of the 17 trials contributing to the primary outcome were rated as high risk of bias because of missing data or imbalanced missingness across intervention and control groups (appendix p 137). However, prespecified sensitivity analyses excluding trials with high risk of bias or high levels of missingness did not change the results. Although BMI Z score is a surrogate measure of clinical adiposity, the measure is an acceptable and commonly used metric for large-scale monitoring of obesity risk^36,37 and is strongly predictive of lifelong obesity.^4–6 Further, other measures of body composition were analysed as secondary outcomes (eg, overweight, obesity, and waist circumference) and findings were consistent with our primary outcome. Only three (<1%) of 6490 children had severe underweight (appendix pp 187–188). We were unable to analyse some prespecified outcomes and subgroups (eg, fidelity of and adherence to interventions and ethnicity) because of excessive heterogeneity in whether and how these outcomes and subgroups were assessed. These challenges highlight the need in future trials for use of core outcome sets, core outcome measurement sets,^83–87 prospective approaches to evidence synthesis,⁸⁸ and a standardised approach for collecting key sociodemographic variables to advance health equity.

Interventions in this study were designed carefully by world-leading experts in the field. Some were grounded in behaviour-change theories, whereas others were based on pilot work and consultation with various stakeholders.²¹ It is surprising and discouraging that we found no evidence of an effect of the interventions on the primary outcome. There are several potential explanations.

First, since the included landmark interventions were designed, new approaches have been developed to identify promising behaviour strategies (eg, stronger application of behaviour-change theory and co-design science)⁸⁹ and target outcomes,²¹ such as targeting more so-called stop behaviours.⁹⁰ Incorporating this knowledge into the development of new interventions could be one promising avenue for future research (eg, in the Greenlight Plus trial,⁹¹ the intervention group had a reduction in BMI Z score of 0·19 [95% CI −0·36 to −0·01], but was not eligible for TOPCHILD because of an active comparator).⁹¹ Importantly, any future research should report results in the context of clinical significance to aid interpretation. However, the interventions included in this study were already quite diverse, and subgroup analyses did not show that any intervention types (eg, setting or dose) worked better than others. Future work will combine detailed intervention deconstruction with individual participant data to further explore potential differences in effectiveness between intervention features.²¹

Second, interventions might be targeting the wrong age group. Although there is a strong theoretical case for starting interventions early, one challenge is that such interventions rely on indirect causal pathways⁹²—ie, the interventions aim to improve child bodyweight status by changing the behaviour of parents, who then need to influence their child’s behaviour. Our consumer panel highlighted that the first year of life can be overwhelming and stressful for parents, leaving them with limited capacity to absorb new content or engage in behavioural interventions. Also, any potential effects on behaviour possibly fade out too quickly to have a lasting effect on child bodyweight status. Once children enter broader social settings beyond the family (eg, early childcare or school), more direct interventions via a settings approach might improve reach and effectiveness.⁹³

Third, the interventions might not have effectively reached priority populations, who are most in need of support. Families experiencing socioeconomic disadvantage and those from minority groups are disproportionately affected by childhood obesity and often face the greatest barriers to accessing interventions.^20,28 This inequity, also referred to as the prevention dilemma,⁹⁴ underscores the need for implementation strategies to address systemic barriers. A TOPCHILD substudy found that although around half of the included interventions targeted priority populations, their success in reaching these groups varied.²⁰ This lack of reach aligns with our findings for key secondary outcomes, which showed that on average outcome values for both intervention and control groups were close to or met recommendations. For instance, on average, children in both groups exceeded minimum physical activity and sleep recommendations. Yet, the observed BMI Z scores were above the population median (ie, >0) in both groups. Notably, we did not find the anticipated differential intervention effects across priority populations, with relevant subgroup analyses (eg, income, education, employment, or immigration status) showing consistent null results. This suggests the interventions were not effective for any population, including priority populations.

Fourth, the included interventions used downstream approaches that require a high degree of individual agency, relying on families to be able and willing to make positive behavioural changes.⁹⁵ Yet, obesity is predominantly driven by upstream environmental and socioeconomic factors that are beyond the capacity of the individual to change.^96,97 Recognising this limitation, the WHO Commission on Ending Childhood Obesity⁴ advocated for a coordinated, whole-system, life-course approach incorporating upstream, low-agency interventions. Such interventions place less onus on individuals and instead focus on transforming obesogenic environments for all populations, thereby mitigating social inequities through addressing disparities in individuals’ capacity to make healthier choices.⁹⁸ Examples of upstream interventions include policies to improve access to healthy foods, providing social support for parents (eg, paid parental leave), increasing green spaces, and regulating unhealthy food marketing.^93,95,96 Supporting the WHO recommendations, a system dynamics model in an Australian context⁹⁹ found that among five intervention scenarios for child obesity prevention, early interventions targeting mothers of children younger than 1 year had the lowest potential effect on reducing obesity prevalence. By contrast, upstream interventions with high coverage over the life course and low individual agency, including a sugar-sweetened beverage tax and child sports vouchers, showed the greatest potential effect. Synergistic effects were observed when combined with setting-based interventions across the life course, including childcare-based and school-based programmes.

Fifth, interventions might need more time to be effective, and measuring outcomes at age 24 months might have been too early. Yet, an individual participant data meta-analysis of four Australasian early obesity prevention trials found no evidence of intervention effects for bodyweight-related outcomes at follow-up at 3·5 years or 5 years of age,¹⁹ making this explanation less plausible. We will further examine BMI Z scores at these later timepoints once more trials have completed follow-up.

This study is the most comprehensive individual participant data meta-analysis in the field to date. Using rigorous, gold-standard methods, we found no evidence that parent-focused obesity prevention interventions affect child BMI Z score at age 24 months (or within 6 months either side) or most of our secondary outcomes. Our findings indicate that existing early, behavioural, parent-focused interventions alone are insufficient to address childhood obesity.

Table 4:

Subgroup analyses by individual-level characteristics for the primary outcome of BMI Z score at age 24 months (±6 months), categorical individual-level subgroups

	Studies	Participants	Floating subgroup-specific intervention effect (95% CI)	Pooled interaction effect (95% CI)
Any formal childcare attendance at age 0–12 months	4	1335
No	4	1156	−0·05 (−0·20 to 0·09)	(reference)
Yes	4	179	0·08 (−0·48 to 0·64)	0·14 (−0·48 to 0·75)
Any formal childcare attendance at age 12–24 months	4	1025
No	4	266	−0·02 (−0·37 to 0·34)	(reference)
Yes	4	759	−0·08 (−0·24 to 0·07)	−0·07 (−0·43 to 0·29)
Partner status	13	4714
In a partnership (married, de facto, or living with partner)	13	4418	−0·01 (−0·08 to 0·07)	(reference)
Single (single, divorced, or widowed)	12	296	−0·20 (−0·48 to 0·07)	−0·19 (−0·48 to 0·09)
Parity or first-time parent	12	5330
First-time parent	12	4221	−0·05 (−0·13 to 0·03)	(reference)
Already has at least one other child	8	1109	0·06 (−0·10 to 0·22)	0·11 (−0·07 to 0·30)
Parent or carer immigration status	7	3636
Primary parent or carer born in trial country	7	2512	0·01 (−0·11 to 0·13)	(reference)
Primary parent or carer born outside trial country	7	1124	−0·02 (−0·21 to 0·18)	−0·03 (−0·30 to 0·25)
Infant sex	17	6505
Male	17	3287	0·01 (−0·07 to 0·10)	(reference)
Female	17	3218	−0·04 (−0·12 to 0·04)	−0·05 (−0·16 to 0·05)
Ambiguous or other	0	0	··	··
Carer education	12	4848
Low education	9	414	0·09 (−0·26 to 0·44)	0·11 (−0·24 to 0·46)
High-school graduate	12	994	0·02 (−0·12 to 0·16)	0·03 (−0·13 to 0·20)
Non-university tertiary education or incomplete university	10	776	−0·04 (−0·18 to 0·11)	−0·02 (−0·19 to 0·16)
University graduate or postgraduate	12	2664	−0·02 (−0·10 to 0·06)	(reference)
Carer employment	15	5598
Any employment (including paid leave)	15	2738	0·03 (−0·07 to 0·12)	(reference)
Unemployed (includes retired, student without employment, unpaid leave, home duties, or charity work)	15	2860	−0·04 (−0·13 to 0·06)	−0·06 (−0·19 to 0·06)

For categorical subgroups, the pooled interaction is the difference in intervention effect between the comparator and reference subgroup (eg, the intervention reduces BMI Z score by 0·19 for those without a partner compared with those who do have a partner). A result is statistically significant if the 95% CI of the pooled interaction effect does not include zero. Floating subgroup-specific intervention effects can be interpreted as the intervention effect that is specific to their respective subgroup (eg, the intervention reduces BMI Z score by 0·01 for those with a partner and by 0·20 for those without a partner. If the pooled interaction effect is not significant, the subgroup-specific intervention effects are not interpreted as statistically different.

Table 5:

Subgroup analyses of trial-level characteristics for the primary outcome of BMI Z score at age 24 months (or within 6 months either side), categorical trial-level characteristics

	Studies	Participants	Test of subgroup differences, χ2	p value
Setting (home, community, or combination)	17	6505	0·57	0·75
In-person delivery (yes or no)	17	6505	0·96	0·33
Intervention mode (individual, group, or both)	17	6505	3·47	0·18
Intervention onset (antenatal or postnatal)	17	6505	0·32	0·57

Data are n, unless otherwise indicated.

Research in context.

Evidence before this study

Although there have been several reviews of trials of childhood obesity prevention, few have focused on infancy, and many are narrative. A Cochrane review of obesity prevention interventions in children aged 0–18 years found that combined diet and physical activity interventions led to a small reduction in BMI Z score in children aged 0–5 years (mean difference −0·07 [95% CI −0·14 to −0·01]). However, this review was limited by reliance on published aggregate data, precluding outcome harmonisation and in-depth subgroup analyses. Updates to this review only examined interventions in age groups older than 2 years. The Early Prevention of Obesity in Children (EPOCH) Collaboration did an individual participant data prospective meta-analysis of four Australian and New Zealand randomised controlled trials. The authors found that parent-focused behavioural interventions for the prevention of early childhood obesity resulted in a small reduction in BMI Z score at age 18–24 months, compared with usual care (−0·12 SDs [95% CI −0·22 to −0·02]). However, this difference was no longer statistically significant when accounting for missing data, and there was insufficient power for subgroup analyses. More studies in a global context were needed to confirm this finding and explore differential effects across key subgroups, including priority populations susceptible to obesity (eg, individuals with lower education or income levels and immigrants), with sufficient statistical power. We searched databases and trial registries (MEDLINE, Embase, CENTRAL, CINAHL, PsycInfo, ClinicalTrials.gov, and WHO International Clinical Trials Registry Platform) from inception up to Sept 30, 2024, without language restrictions, to identify randomised controlled trials investigating the effectiveness of obesity prevention interventions commencing antenatally or during infancy. The search strategy included terms related to weight (eg, “obes*”, “overweight”, “body mass index”, and “adiposity”); behavioural interventions (eg, “behavio?r*”, “diet*”, “physical activity”, “sleep”, and “feeding”); children, parents, and families (eg, “child”, “infan*”, “pregnan*”, “prenatal”, and “parent$”); and study design (eg, “randomi#ed”).

Added value of this study

This is the largest systematic review with an individual participant data meta-analysis in the field to date, including 31 trials set across ten countries with 28 825 participants. Of 18 eligible completed trials that assessed our primary outcome (BMI Z score at age 24 months [±6 months]), we obtained individual participant data for 17 (94%), including 9128 participants (97% of 9383 eligible participants). We used rigorous data checking, harmonisation, and integrity and risk-of-bias assessments enhanced by the availability of individual participant data and collaboration with trial representatives to create a global dataset. Access to original data enhanced statistical power and enabled advanced and complex analyses, including examination of important obesity-susceptible subgroups. The resulting high-quality TOPCHILD dataset provided robust evidence to address our research questions. Here we report findings for the primary outcome, BMI Z score, and for 33 prespecified secondary anthropometric and behavioural outcomes.

Implications of all the available evidence

This large individual participant data meta-analysis provides evidence that existing approaches to parent-focused behavioural interventions delivered up to 12 months of age are insufficient to affect BMI Z score at age 24 months (or within 6 months either side) or key obesity-related behavioural outcomes covering diet, feeding, physical activity, sleep, and parenting. These findings underscore the need to rethink current behavioural approaches to prevent obesity in early childhood.

Data sharing

The TOPCHILD Collaboration has obtained permission to use but does not own all data used in these analyses. Where possible, individual participant data collected for this study, including a data dictionary, will be made available following a moderated access process, whereby a proposal needs to be approved by the original data custodians (ie, the trial investigators) and a cross-institutional data access agreement needs to be signed. The statistical analysis plan (appendix pp 8–119) and protocol are already publicly available. Please contact the corresponding author or the TOPCHILD Collaboration (topchild.study@sydney.edu.au) to request data access.