Early childhood (birth through 5 years) is recognized as a critical period during which key health behaviors (diet, physical activity, sedentary behavior, and sleep) are established. However, evidence shows that these behaviors are suboptimal from early life. Despite the well-established benefits of breastfeeding, fewer than half of infants are exclusively breastfed for the first 6 months [1]. Similarly, global evidence indicates that adherence to early childhood dietary [2-4] and movement behavior guidelines is low, with large international reviews reporting that only a small proportion of children meet recommendations across physical activity, sedentary behavior, screen time, and sleep [2,3]. Of particular concern is the fact that these suboptimal behaviors can track into later childhood and adolescence [4,5], underscoring the need for interventions to promote health behaviors from a young age. Recent studies have also highlighted widening socioeconomic inequalities in children’s early environments [6,7] and less optimal diet, physical activity, sedentary behavior, screen time, and sleep among children from socioeconomically disadvantaged backgrounds [8-11], reinforcing the need for accessible intervention strategies for families who may have limited access to traditional face-to-face services.

Existing early childhood interventions have shown varied success in improving diet, physical activity, sedentary behavior, and sleep habits [12-15]. Traditionally, these interventions have relied largely on time-consuming and costly face-to-face delivery, with limited consideration of scalability or implementation at scale [16]. In contrast, digital interventions (eHealth and mobile health [mHealth]) have the advantage that they can be delivered anywhere, anytime, maximizing potential reach across diverse socioeconomic, geographical, and cultural backgrounds. Recent years have seen a rapid expansion of digital health solutions, including web-based platforms, mobile apps, and wearable technologies [17], supported by growing evidence of their potential for scalability and cost-effectiveness [18,19].

Reflecting this broader growth in digital health, there has also been a marked increase in digital interventions targeting diet and movement behaviors across all age groups, with most reporting efficacy in changing behavior [20-22]. This also includes a growing interest in the feasibility and effectiveness of these types of interventions for targeting childhood obesity and obesity-related behaviors [23-31]. However, these reviews have largely focused on childhood broadly (0‐18 years) [23,24,27-30] or have focused solely on preschoolers (3‐6 years) [25,26]. Many have also examined single behavioral domains such as physical activity or sedentary behavior [27-29], or specific population groups such as Indigenous mothers of young children [31]. In addition, most evaluate digital interventions that involve some degree of human support or multicomponent programs and do not address key design and implementation factors. As such, they offer limited insight into early childhood as a distinct developmental period or the potential of autonomously delivered interventions to support parents across multiple behaviors in the first 2000 days. Given that health behaviors are largely shaped early in life and that this life stage is characterized by unique developmental and parental influence, a review focusing solely on this period is warranted.

Moreover, co-design, engagement, and implementation factors are highly important for successful intervention delivery [32,33]; yet, these elements remain inconsistently assessed and underreported in interventions targeting parents of young children [34]. Thus, this review provides a timely and comprehensive synthesis of autonomously delivered digital interventions targeting parents across the first 2000 days (conception to age 5 years), examining multiple behavioral domains (breastfeeding, feeding practices, diet, physical activity, sedentary behavior, screen time, and sleep) and integrating evidence on co-design, process evaluation, and engagement. This broader scope allows us to identify key gaps, emerging patterns and implications for the design and implementation of scalable digital strategies in early childhood.

The primary aim of this systematic review was to examine whether autonomously delivered digital interventions targeting parents are effective at increasing physical activity, reducing sedentary behavior, improving nutrition (breastfeeding, feeding practices, and dietary outcomes), and/or optimizing sleep among children aged 0‐5 years. The secondary aim was to review the reporting of co-design practices, user engagement, and process evaluation, and to assess how engagement influences intervention effectiveness.

We included randomized controlled trials (RCTs) of interventions delivered to parents and caregivers of children in the first 2000 days of life (herein referred to as parents) solely via digital technology (mHealth/eHealth). Interventions needed to aim to improve one or more of the following behaviors: increase physical activity, reduce sedentary behavior, improve nutrition (breastfeeding, food intake, and feeding practices), and/or optimize sleep among children.

We only included interventions that solely used digital technologies to autonomously deliver the intervention (ie, where no personnel were needed to deliver and/or maintain the intervention). We used this definition as we were interested in solutions for intervention delivery that could be more easily and cost-effectively scaled compared to interventions requiring delivery personnel (with or without a digital component). Therefore, we excluded digital interventions that required delivery personnel for one-to-one support, for example, telephone coaching calls or face-to-face counseling sessions with a supplementary online social support group.

We considered direct outcome measures of children’s target behaviors (eg, parent-reported or accelerometer-measured physical activity) and indirect outcome measures known to influence children’s target behaviors (eg, parental feeding style or changes to food environment). There were 2 key reasons we decided to also include indirect outcome measures: (1) it can be difficult to accurately measure child behaviors, particularly for younger children (eg, breastmilk intake), and (2) national guidelines for infant feeding [35] and movement behaviors in the early years [36] include parental influences known to impact child behaviors (eg, for establishing healthy sleep habits, parents can set up a calming bedtime routine and consistent sleep and wake-up times).

Studies were also excluded if they were (1) not an RCT design, (2) solely targeting other caregivers (eg, grandparents or childcare providers), (3) among parents with older children (≥6 years old), (4) among parents with children who had clinical health conditions (eg, diabetes and premature birth), and (5) interventions delivered primarily within the antenatal period (noting that those delivered from pregnancy to infancy were included). We limited studies to primary research published in English in the peer-reviewed literature. Literature reviews and meta-analyses, theses, conference proceedings, and gray literature were not included.

We searched 7 electronic databases, including Embase (Elsevier), Academic Search Complete (EBSCO), CINAHL Complete (EBSCO), Global Health, MEDLINE Complete (EBSCO), PsycINFO (EBSCO), and SPORTDiscus (EBSCO). All databases were searched individually rather than simultaneously via a multidatabase platform. We did not search study registries, conference proceedings, websites, or other online sources, nor did we undertake citation searching or contact authors or experts. No supplementary search methods beyond database searching were used, as the review was limited to peer-reviewed RCTs. Searches were conducted in December 2022 and updated in August 2024 and again in January 2026.

This systematic review was prospectively registered with PROSPERO (International Prospective Register of Systematic Reviews; ID: CRD42022372639) [37]. We made 2 amendments to the registered protocol: (1) to limit study design to RCTs due to the recent increase in publications and (2) the use of a different quality appraisal tool specific to RCTs. The PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analysis) 2020 statement guidelines [38] and the PRISMA-S (Preferred Reporting Items for Systematic Reviews and Meta-Analyses Literature Search Extension; Checklist 1) extension for reporting search strategies were followed to ensure transparent reporting of the search process [39].

The search strategy was developed by BM, SM, and KD, in consultation with all authors. We ran preliminary searches and sought technical guidance from Deakin University librarians to refine the search strategy. The search strategy included a combination of keywords to capture concepts according to the Population, Intervention, Comparison, Outcomes, and Study design (PICOS) tool: (1) children aged ≤5 years, (2) mHealth or eHealth intervention, (3) intervention behavioral targets (ie, child physical activity, sedentary behavior, nutrition, and/or sleep), and (4) study design (see Multimedia Appendix 1). The search strategy was not adapted from previous reviews and did not undergo a formal peer-review process.

All search results were exported to Covidence and duplicates were removed. Two reviewers (SM or JS and either BM, KD, or KH) independently screened all articles first by the title and abstract; a third independent reviewer resolved discrepancies (BM or KD). Two reviewers screened the full texts (BM and SM or JS and KD) using the inclusion/exclusion criteria described above. Discrepancies were resolved by a third reviewer (KD or KH). During the updated review in 2026, 2 researchers, CS and SR, assisted with screening, data extraction, and evaluation (disagreements were resolved by JS) following the same procedure.

Two reviewers (JS and BM or CS and SR) extracted the following information using a prepiloted data collection template developed for this review. Template fields included study characteristics (eg, authors, year of publication, country, study design, study aims, and target behaviors), setting and participants (eg, setting, inclusion/exclusion criteria, recruitment, and sample size), intervention description (eg, technology used, delivery mode, content, onboarding processes, and engagement), and participant outcomes (eg, measures related to target behavioral outcomes, data collection tool and method, and results). Co-design (eg, stakeholder engagement in intervention development), intervention theory (ie, the use of a behavior change theory in intervention development), process evaluation outcomes (eg, acceptability, feasibility, and reach), and intervention engagement data (eg, app analytics, self-reported use, and impact of engagement on intervention effectiveness) were extracted using a combination of the data extraction tool at Elicit.com [40] (Ought; an online platform that automates data extraction) and manual extraction and checked for accuracy and completeness by one author (KD or CS).

We extracted data on all relevant behavioral outcomes (breastfeeding, feeding practices, diet, physical activity, sedentary behavior, screen time, and sleep). For each outcome, all reported time points and measurement tools were collected when available. We also extracted additional study variables, including participant characteristics, intervention features, delivery mode, co-design processes, theoretical underpinnings, process evaluation outcomes, and engagement metrics. When information was unclear or missing, assumptions were not made; instead, data were recorded as reported.

The risk of bias and quality assessment for each individual study was assessed by 2 authors independently (JS and KD or CS and SR) using the Joanna Briggs Institute (JBI) Critical Appraisal Checklist for RCTs [41]. The checklist includes 13 items, where each item can be scored yes, no, unclear, or not applicable, in the categories of selection and allocation; administration of intervention/exposure; assessment, detection, and measurement of outcomes; participant retention; and statistical conclusion validity. As the interventions were autonomously delivered (eg, via digital platforms), there were no treatment deliverers involved; therefore, the item “Treatment deliverers blinded” was not applicable in this context. The initial interrater agreement between JS and KD was 81%, and 85% for CS and SR. Discrepancies in assessment between authors were discussed (JS and KD or CS and SR) until consensus was reached.

Consistent with Cochrane guidance [42], due to the heterogeneity in definition, measurement, and reporting of outcomes across studies, meaning they did not estimate the same underlying effect, a meta-analysis was not able to be conducted. Therefore, following PRISMA [38] and synthesis without meta-analysis recommendations [43], a structured narrative synthesis was undertaken, grouping studies by target age group (pregnancy-infancy, toddlers, and preschoolers) and by behavioral domain (breastfeeding, diet, physical activity, sedentary behavior, and sleep). We summarized the direction of effect and patterns in effectiveness using structured tables.

We did not conduct a formal certainty-of-evidence assessment because of the heterogeneity of the included studies and the lack of commensurable effect estimates. Instead, we considered study-level risk of bias and the consistency of findings when interpreting the results.

The literature search yielded 14,352 unique articles. Most records excluded at the title and abstract stage were due to at least one of the following reasons: the study did not examine an autonomously delivered digital intervention (eg, clinical-delivered or face-to-face), targeted populations outside the scope of the review (eg, older children, adolescents, or adults), or did not use an RCT design (eg, observational studies, qualitative studies, pilot feasibility work without randomization, or large noninterventional epidemiological analyses). Following screening, 258 full-text reports were assessed for eligibility, of which 49 studies describing 38 interventions were included in this review. No studies were excluded at full-text review that appeared to meet the inclusion criteria; all exclusions were based on predefined criteria (eg, wrong age group, design, and intervention type). The screening process is presented in Figure 1.

An overview of the targeted outcomes and age groups for the 38 included interventions [44-81] is presented in Table 1. Almost half of the included studies (50%, 19/38) [45,48,51,53,55,56,58,59,62,64,65,67,69,70,72,76,77,79,80] focused on improving breastfeeding, 26% (10/38) [44,46,47,49,52,60,61,71,73,81] focused on multiple behaviors, 16% (6/38) [50,54,66,74,75,78] on diet only, 5% (2/38) [63,68] on sleep only, and 3% (1/38) [57] on physical activity only. The multiple behavior interventions targeted different combinations of breastfeeding, feeding practices, diet, physical activity, sedentary behavior, screen time, and sleep. Only one study focused on all behaviors (breastfeeding, feeding practices, physical activity, sedentary behavior, screen time, and sleep) [61], and no study focused solely on sedentary behavior/screen time. One study included children aged 0‐3 years but is reported under the infant age group because the mean child age fell within infancy [72].

Tables 2-4 present the study characteristics for the included studies divided by age group. In summary, the included studies comprised RCTs (n=32) [44-63,68,70-80], pilot RCTs (n=4) [47,64-66], one feasibility RCT [67], and one pilot study using a micro-RCT design [81]. Studies were published between 2011 and 2026, with most studies being published since 2020 (27/38, 71%) [51-62,64,65,69-81]. Most studies were conducted in the United States (n=14) [44,45,47,48,57,64-70,76], followed by Australia (n=5) [49,52,53,71,80], China (n=3) [51,61,72], Sweden (n=2) [46,60], Norway (n=2) [50,54], Ethiopia (n=2) [77,79], and Thailand (n=2) [73,74]. Single studies were conducted in India [56], Myanmar [59], Nepal [78], Iran [75], Turkey [58], Vietnam [55], Spain [62] and Canada [81]. The studies included sample sizes ranging from 18 [81] to 5095 [56].

The susceptibility to bias of the included studies is presented in Table 2. Briefly, all studies measured outcomes the same way for groups and had an appropriate study design (n=38) [44-81]. Most studies used appropriate statistical analyses (n=37) [44-67,69-81], had true randomization (n=36) [44-46,48-67,69-81], had complete follow-up (or, if not, adequately described and analyzed differences between groups in terms of their follow-up; n=31) [44-64,66,67,69,73-79], used intention-to-treat analysis (n=30) [44-62,64,65,67,69,71-73,77-80], and treated groups identically (other than the intervention of interest; n=33) [45-48,50-56,58-60,63-81]. Findings were mixed for other risk of bias items; in particular, concealment of group allocation (yes: n=17 [46,48-51,55,56,59,62,64,66,69,71,74,75,78,80]; unclear: n=18 [44,45,47,52,54,57,58,63,65,67,68,70,72,73,76,77,79,81]; no: n=3 [53,60,61]), similarity of groups at baseline (yes=25 [44-47,49,50,52-58,60,65,67,69,70,73-75,77-80]; no=11 [48,51,59,61-64,66,68,71,72]; unclear=2 [76,81]), blinding of participants (yes=6 [44,49,55,68,69,80]; no=27 [45-48,50,52-54,56-61,64,65,70-79,81]; unclear=5 [51,62,63,66,67]), blinding of outcome assessors (yes=22 [44,46-49,51,52,55-57,59,60,62,64,68,69,73-75,77-79]; no=3 [61,66,71]; unclear=13 [45,50,53,54,58,63,65,67,70,72,76,80,81]), reliability of outcome measures (yes=28 [44-50,52-55,57,60,65-69,71,72,74-81]; unclear=10 [51,56,58,59,61-64,70-73]). The item assessing the blinding of treatment deliverers was not applicable for many studies, as the digital interventions under investigation were all autonomously delivered.

Co-design, intervention theory, process evaluation, and engagement outcomes for studies focusing on pregnancy to infancy, toddlers, and preschoolers are shown in Tables S8, S9, and S10 in Multimedia Appendix 2, respectively.

This systematic review provides a comprehensive overview of autonomously delivered digital interventions targeting multiple behaviors in the first 2000 days and examines co-design, engagement, and process evaluation, areas rarely assessed in previous reviews. Thirty-eight interventions were included in the review, with most focusing on improving breastfeeding practices by targeting mothers and their youngest children (newborns and infants) and a growing number of studies targeting toddlers and preschoolers. Overall, intervention designs varied considerably, and results were mixed for all targeted age groups with no apparent trend in intervention characteristics (eg, target behavior and digital delivery mode) for interventions shown to be effective. Although most studies reported some form of co-design or end-user engagement, very few examined the impact of engagement on the efficacy of the intervention.

The variability in intervention design is not unique for autonomously delivered digital interventions, and previous reviews on digital interventions for improving breastfeeding have similarly reported heterogeneity; for example, in delivery modes [82-84]. In terms of effectiveness, most of the included studies in our review reported no differences in breastfeeding outcomes between the groups and reported mixed results. These findings are similar to previous systematic reviews and meta-analyses focusing solely on mobile apps [83], remote provision of breastfeeding support education (eg, telephone, SMS text messaging, social media, video call, and email) [82], or mHealth-based interventions to promote breastfeeding [84]. To illustrate, Ziebart et al [83] found insufficient evidence for sustained beneficial effects of breastfeeding promotion and support through mobile apps on breastfeeding rates, while Gavine et al [82] concluded that remote interventions can be effective for improving exclusive breastfeeding at 3 months but with low certainty of evidence due to risk of bias, substantial heterogeneity, and imprecision in some outcomes. In contrast, Qian et al [84] reported, amongst other things, improvements in exclusive breastfeeding rates up to 6 months after delivery in comparison with usual care. These previous reviews included only up to 4 of the studies captured in our review, likely due to differences in search timeframes and inclusion/exclusion criteria. For example, Ziebart et al [83] focused exclusively on mobile apps related to breastfeeding and excluded web-based interventions; Gavine et al [82] limited their search to studies published after 2010 and focused on remote care broadly; and Qian et al [84] included a wider range of mHealth modalities (eg, phone calls, SMS text messaging, and interactive systems) but only considered breastfeeding outcomes. In contrast, our review included RCTs targeting a broader set of health behaviors (breastfeeding, feeding practices, diet, physical activity, sedentary behavior, and sleep) in children aged 0‐5 years, delivered solely via autonomous digital technologies. Thus, our review provides an important addition to the existing evidence.

Moreover, we report the effectiveness of interventions covering the first 5 years of life, which is recognized as a critical period for establishment of health behaviors. The studies targeting toddlers showed promising results for diet, screen time, and sleep but not physical activity. However, there were only 6 [60,68,71,73,74,78] studies in this age group, highlighting the need for more research targeting toddlers. Similarly, the number of studies targeting preschoolers was limited, and results indicated significant improvements for feeding practices and diet, mixed findings for screen time, and no overall differences for children’s physical activity, sedentary behavior, and sleep, although one trial reported context-specific reductions in children’s sedentary time and increased parental moderate-to-vigorous physical activity. Previous systematic reviews on digital interventions to promote these health behaviors in preschoolers have also reported mixed findings [25,26]. For example, Zhou et al [26] reported significant improvements for dietary behaviors and sleep but no significant improvements in physical activity for parent-based eHealth interventions. In contrast, a systematic review and meta-analysis on the effectiveness of eHealth interventions for promoting 24-hour movement behaviors in preschoolers [25] reported small but positive effects on physical activity, sedentary time, and sleep. One potential explanation for the discrepancies in results could be differences in intervention delivery, as most of the included studies in the meta-analysis relied on human interaction (eg, courses with staff members, motivational coaching, individual discussion/telephone calls, face-to-face workshop, and home visits) [25,26]. Although autonomously delivered interventions have greater potential for scalability and potential cost-effectiveness, digital interventions including a delivery personnel component might be more effective; however, considering the paucity of studies in this age group, more research is required to determine this.

Reporting of elements related to the design and evaluation of interventions, including co-design and process evaluation, is important considering that these can help improve intervention effectiveness [85] and identify key components that contribute to the success of interventions [86], respectively. Another important aspect is participant engagement, as it can provide information on levels of intervention exposure and uptake, which are crucial for effectiveness, as well as help identify factors that may optimize engagement [33] and ultimately inform the development of more impactful interventions. Although most of the included studies reported some form of co-design or end-user engagement, only 6 studies [56,62,69,71,72,78] examined the impact of engagement on intervention effectiveness. In contrast with previous findings in other populations (eg, [87,88]) and outcomes such as parenting practices and cognitions [71], the studies included in this review reported mixed findings across behaviors. All breastfeeding-focused interventions reported that higher engagement was not associated with intervention effectiveness on the targeted outcomes (ie, breastfeeding) [56,62,69], while studies targeting breastfeeding and feeding practices [72], diet [78], and movement behaviors [71] found that higher engagement was associated with more favorable outcomes. Nevertheless, the low reporting rate of engagement suggests that the results from this review about the impact may not fully reflect the intended interventions. Although engagement data are supposedly easy to collect in digital interventions (eg, apps), previous research has highlighted a gap in evaluating and reporting how engagement influences intervention effectiveness [89]. This raises critical questions, such as why engagement data are often omitted, and whether low engagement levels could be a contributing factor, with researchers hesitant to report underwhelming results. Despite the assumption that digital interventions increase reach, this may not translate to increased engagement [34]. Ultimately, understanding whether and how engagement mediates outcomes is essential to determine the true value of these interventions, and addressing these gaps is crucial for ensuring that scalable interventions are also impactful and meaningfully engaged with by users. This review highlights key considerations for improving future research, including the evaluation and reporting of the impact of engagement on the effectiveness of digital interventions for promoting health behaviors in early childhood.

This review has several strengths and limitations. A key strength was the systematic approach used to search, screen, and synthesize the literature, including the PROSPERO registration of the review protocol and the use of the JBI Critical Appraisal Checklist for RCTs. Moreover, we focused on digital interventions that were delivered autonomously and thus, in theory, have the capacity to be scaled up and delivered at large without heavy researcher or staff input. Another strength is that we also considered intervention development (co-design and intervention theory) and participant engagement, which are important elements for intervention effectiveness [85]; however, this broader evaluation was constrained by the extent of reporting within individual studies. The present review also has limitations to acknowledge. First, despite comprehensive searches across multiple databases, it is possible that relevant studies were missed, particularly unpublished or non-English studies. Second, we restricted the review to autonomously delivered digital interventions, which limit our findings to hybrid models that include human support. Third, considerable heterogeneity in study populations, intervention content, delivery formats, and outcome measures prevented and limited our ability to perform a meta-analysis. The wide variety in intervention objectives, settings, methodologies, and delivery modes also made it difficult to compare findings across studies. Finally, digital health interventions evolve rapidly, and more recent innovations may not yet be represented in the current evidence base, as highlighted by recent evidence underscoring the challenges of synthesizing findings in this fast-moving field [22].

In terms of strengths and limitations for the individual studies, although most studies had an appropriate study design and use of statistical methods, findings were mixed for other risk of bias items, including concealment of group allocation, blinding of participants, and outcome assessors, as well as reliability of outcome measures. Considering the nature of the interventions, blinding of participants and researchers might not be feasible, suggesting that checklists assessing study quality specific to digital interventions are warranted. Most studies also used subjective methods to assess outcomes, which are inherently subject to misreporting biases. Finally, most studies were conducted in high-income countries, which limits generalizability to low-income countries.

This review shows that autonomously delivered digital interventions for early childhood are highly heterogeneous and demonstrate mixed effectiveness, making it difficult to identify which components are most impactful. It is innovative in synthesizing evidence across the first 2000 days while simultaneously examining co-design, engagement, and implementation factors, dimensions rarely brought together in previous work. Unlike earlier reviews that focus on older children, single behaviors, or interventions involving human support, this review focuses solely on scalable autonomously delivered digital interventions for children 0‐5 years, a formative period for long-term health. Most importantly, it adds new insight by identifying three priority evidence gaps: (1) the scarcity of studies targeting toddlers and preschoolers, (2) inconsistent and incomplete reporting of engagement, and (3) limited understanding of how engagement influences outcomes. While autonomous digital interventions offer clear advantages in reach and scalability, their usefulness ultimately depends on whether interventions remain engaging, relevant, and effective for families. Together, these findings define priority areas for future research and clarify what is needed to strengthen the evidence base for scalable digital interventions in early childhood.