A protocol for this systematic review was prospectively published in PROSPERO (CRD42023492644) [30] and included a variety of outcomes related to physical activity, diet, and sedentary behavior; however, this paper reports on the breastfeeding outcomes only. This systematic review and meta-analyses were conducted as per methods outlined in the Cochrane Handbook [31] and reported as per the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines [32] (Checklist 1) as well as the PRISMA-S (Preferred Reporting Items for Systematic Reviews and Meta-Analyses-Searches) checklist [33] (Checklist 2), and PRISMA checklist for Abstracts (Checklist 3).

Search terms for “Digital Health Interventions” were adapted from previous systematic reviews to suit our research question [39-41] and reviewed by an experienced research librarian.

The search was based on broad domains using MeSH related to “Breastfeeding” combined with “Parents/carers” or “Infant/Newborn” combined with “Digital Health Intervention.” Cochrane recommended filters were applied to help identify RCTs, and limits were applied to the database searches (where possible) to identify human studies [42]. The search terms for each electronic database are listed in Multimedia Appendix 1.

Following deduplication of identified titles and abstracts using the “find duplicate” function in EndNote (Clarivate; matching references for author, year, title, and reference type), additional duplicates were identified and removed within Covidence. Further, non-RCTs were removed prior to title and abstract screening using Covidence artificial intelligence filters [43].

Following deduplication of references, pairs of review authors (JJ, AR, ALB, RL, KP, KS, SS, and TD) independently screened titles and abstracts using Covidence software. Where discrepancies arose or where there was insufficient detail to determine eligibility based on the title or abstract, studies were advanced to full-text review.

Pairs of review authors (JJ, ALB, BS, RL, and NH) independently assessed full-text papers for eligibility using Covidence. Where discrepancies between reviewers for study inclusion could not be clearly reconciled by the lead reviewer (JJ), a field expert (RS) was consulted as a third reviewer to determine final study inclusion.

Study characteristics were extracted by 1 author (SK) and verified by the lead reviewer (JJ), using an adapted and piloted version of the Cochrane Public Health data extraction template. Extracted data included author, location, design, population, and intervention description (as per the TIDieR [Template for Intervention Description and Replication] checklist [44]) and comparator components, duration, setting, sample size, participants’ age, outcomes, and results.

Primary outcome data were extracted independently by the lead reviewer (JJ) and checked by a second reviewer (SK). Any discrepancies were resolved by consensus.

Pairs of unblinded review authors (JJ, SK, AR, RL, and LST) independently assessed the risk of bias of included studies using the Cochrane Risk of Bias 2 tool for randomized trials [50]. Additionally, as per the Cochrane Handbook, the risk of bias assessment focused on the main outcome of the review (breastfeeding status) and assumed an intention-to-treat effect (ie, based on assignment to the intervention) [51].

All recommended domains were assessed including (1) randomization process, (2) deviations from intended interventions, (3) missing outcome data, (4) measurement of the outcome, and (5) selection of the reported result. If the study was a cluster RCT, additional relevant criteria were assessed relating to the timing of identification or recruitment of participants. Overall study risk of bias was based on the algorithm recommended in the Cochrane Handbook [51]. For example, studies were considered high risk of bias overall if one or more risk of bias domains were assessed as high risk.

The certainty of evidence related to our primary outcomes was evaluated using the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) approach to categorize the certainty of evidence as “high,” “moderate,” “low,” or “very low.” Two authors (JJ and SK) used the GRADEpro GDT software (Evidence Prime Inc) to conduct separate summaries of findings and assessments of the level of evidence [52]. As part of the GRADE assessment, evidence was evaluated based on risk of bias, inconsistency, indirectness, imprecision, and other biases including publication bias and small study effects (by examining funnel plot asymmetry) [53]. Any discrepancies between author assessments were resolved by consensus.

The search identified a total of 11,106 records that underwent title and abstract screening (see Figure 1 for PRISMA flow diagram). Of these, 254 full-text papers were reviewed for eligibility, and a total of 208 records were excluded at full-text screening (primarily due to studies not reporting on the primary outcome: n=66, wrong study design: n=49, or wrong intervention: n=45). A total of 46 studies (all published in peer-reviewed journals) met the review eligibility criteria and were included within the review.

Within the 46 included studies, there were 96 trial arms and 33,785 participants. Key details of the studies including study design, country, details of the DHI, intervention and comparator participants, outcomes, time points of measurement, and method of synthesis are included in Table 1.

In total, 11 of the included studies were conducted within the United States [56,59,61,62,67,68,71,75,81,91,92], 7 studies were conducted in Turkey [55,58,60,63,65,87,90], 5 in China [70,77,80,98,99], 4 in India [78,79,88,89], 3 in Iran [57,66,94], and 2 studies each were conducted in Australia [93,97], Spain [73,96], and Uganda [84,85]. The remainder of studies were conducted across a broad cross-section of countries, with 1 each conducted in Brazil [64], Canada [54], Ecuador [82], Ethiopia [72], Israel [83], Kenya [95], Myanmar [76], Nigeria [86], Somalia [74], and Vietnam [69].

In total, 35 of the included studies were RCTs [54-65,67-71,73,75-77,79,81-83,85-87,89-92,96,98,99], 4 were multiple intervention arm RCTs [66,93,95,97], and 7 were cluster RCTs [72,74,78,80,84,88,94]. Trial populations ranged in size from 33 [70] to 13,491 participants [78].

The majority (38/46) of the included studies targeted only mothers, recruited either prenatally (n=23) [57,59,60,62,63,68-72,75,76,81,84,85,88-90,92,95-97,99] or postnatally (n=15) [55,56,58,61,64-66,73,82,83,86,87,91,94,98]. In total, 7 of the studies targeted both parents [54,67,74,77-80], including 2 that targeted households (eg, primary caregivers of young children broadly) [74,80]. Only 1 intervention was “father-focused” [93].

Included interventions were summarized in Multimedia Appendix 2 [54-99], as per the TIDieR checklist [44]. Most (35/46) of the included studies had a focus on improving breastfeeding behaviors. The remaining 11 studies included breastfeeding content but had a focus on maternal and infant health more broadly [61,74,76,79,80,82-85,95,97]. In total, 25 of the interventions were delivered using mHealth (eg, messages or support delivered via mobile or smartphone) [55,57-59,61,62,67-70,72,74,76-78,80,81,84,85,87,89,93,95,96,99], 8 were delivered via computer-based eHealth [54,56,63,64,71,73,90,91], and 4 were delivered via telehealth [79,82,86,92]. A further 7 interventions combined mHealth and telehealth strategies [65,66,75,83,88,94,97], and 2 combined computer-based eHealth and telehealth strategies [60,98]. Interventions were delivered for a duration ranging from 1 week (posthospital discharge [66]) up to 12 months post partum [67,68,79]. In total, 13 of the interventions were delivered via automated systems [61,62,67,69,74,76,79,85,90,91,93,96,99], while the remaining 33 interventions contained some elements that required delivery via trained or qualified personnel (including peer counselors [70,73,75,81,84], research staff [54,55,57,58,60,63,65,66,72,87], academics [64], psychologists [71], nurses [82,95,97], lactation specialists [56,77,83,88,98], gynecologists [59], and pediatricians [86,89,94]).

In total, 11 of the studies indicated that the intervention content or information was underpinned by a theory or framework, the most common being the health belief model (n=3) [67,76,97] and the theory of planned behavior (n=3) [62,75,94]. Other frameworks and models included (1) breastfeeding coparenting framework [54], (2) self-regulation model [56], (3) social cognitive theory [62], (4) social learning theory [71], (5) health action process approach to behavior change [81], (6) Pender’s health promotion model [90], (7) Dennis’ breastfeeding self-efficacy framework [57,98], and (8) behavior change communications framework [72]. Coding to the SEM showed that all interventions targeted individual factors (eg, knowledge, self-efficacy, motivation, and skills), 17 also included components targeting interpersonal factors (eg, coparent support) [54,62,64-66,69-71,73-75,78,81,84,93,94,98], and 20 included community factors (eg, linking to health services or professionals) [56,59-61,68,69,71-73,77,78,82,83,85,86,88,89,94,95,98]. A total of 21 of the interventions incorporated elements that were tailored, including (1) participant preferences for time or day of receiving content (n=1) [95], (2) participant preferred language (n=6) [62,74,78,81,88,96], (3) the age and stage of gestation and/or the infant (n=3) [76,85,99], (4) the delivery pathway (n=1) [61], and (5) based on specific questions or issues raised by the participant (n=11) [56,59,60,65,66,70,71,73,82,83,96].

Included studies compared the intervention to either nondigital usual care control (n=35), attention control (n=9) [59,62,64,69,71,74,76,97], or nonintervention control (n=2) [63,79].

All measures of the primary outcome (EBF, any breastfeeding, and breastfeeding duration) were measured via self-reported methods. EBF was reported in 41 of the included studies [54-60,62-66,68-93,95,97-99], and “any” breastfeeding was reported in 22 studies [57,59,61,62,67,68,70,73,76-81,83,92-94,96-99] (Table 1).

A total of 39 studies were included in the EBF meta-analysis [54-60,62-65,68-73,75-79,81,82,84-93,95,97-99], 21 were included in the meta-analysis exploring “any” breastfeeding rates [57,59,61,62,67,70,73,76-81,83,92,93,96-99], and 7 were included in the meta-analysis for breastfeeding duration [64,66-68,77,91,98].

Breastfeeding self-efficacy was reported in 14 included studies, 2 of these studies used the long-form version of the validated tool [62,66], while 12 used the short-form version [54,57,58,63,65,68,70,76,89,91,93,98] of the Breastfeeding Self-Efficacy Scale [100]. The long-form of the Breastfeeding Self-Efficacy Scale uses a 33-item tool for scoring breastfeeding self-efficacy [101], and the short-form includes 14 items [102]. In both cases, a higher score indicates higher levels of breastfeeding self-efficacy. As such, all 14 studies were included in the meta-analysis [54,57,58,62,63,65,66,68,70,76,89,91,93,98].

Patel et al [88] was the only included study to report on intervention costs (ie, costs incurred by the health care providers and patients) or cost-effectiveness (ie, incremental cost-effectiveness).

Adverse outcomes were mentioned in 5 of the included studies and related to adverse events as reported by study participants [62,68], adverse infant feeding practices [76], adverse clinical nutrition status [86], and serious adverse events as a result of the intervention [95].

For the 39 RCTs, overall risk of bias was mostly assessed as having “some concerns” (n=15) [54,56,57,62,64,66,67,70,71,86,87,91,92,95,99] or “high risk” (n=20) [55,58-61,63,65,68,69,73,75,77,79,81-83,85,89,90,98] of bias. “Some concerns” for study risk of bias was most frequently indicated for domain 5 “the selection of reported results” due to the absence of a preregistered data analysis plan. “Some concerns” for study risk of bias was also frequently indicated for domain 1 the “randomization process,” due to the detection of baseline differences between intervention groups. “High risk” of bias was most frequently indicated for domain 2 “deviations from intended interventions” and domain 3 “missing outcome data,” whereby breastfeeding data were not available for “almost all” participants (Figures 2 and 3).

For the 7 cluster RCTs, 1 study [74] was “low risk” of bias overall, and the other 6 were “high risk” of bias [72,78,80,84,88,94]. “High risk” of bias was most frequently indicated for domain 2 “deviations from the intended interventions” and domain 4 “measurement of the outcome” (Figure 4).

This systematic review and meta-analysis included a total of 39 RCTs and 7 cluster RCTs exploring the effect of parent-directed DHIs (inclusive of eHealth, mHealth, and telehealth interventions) on breastfeeding behaviors, spanning broad contexts and populations from high- to low-income countries. Using gold-standard systematic review methodology, this review provides global evidence exploring the impact of DHIs on key breastfeeding behaviors (ie, EBF and breastfeeding duration), addressing a timely and policy-relevant evidence gap. Our pooled findings indicate that the odds of EBF are approximately double for those receiving a DHI than control (OR 2.58, 95% CI 1.91-3.50; P<.00001); however, our certainty of this evidence is very low. Our pooled analysis also suggests with moderate certainty an effect for parent-targeted DHIs on breastfeeding duration compared with control (SMD 0.48; P=.0007), equivalent to an additional 2 weeks of breastfeeding.

The results from this review are aligned with previous reviews [27-29,103], indicating that parent-targeted DHIs are effective for improving breastfeeding behaviors, despite extending the literature search to include high-, middle-, and low-income countries, while examining a comprehensive range of modalities for DHI delivery. The results of this review are highly encouraging, given that small population-level increases in breastfeeding exclusivity and duration contribute significant and ongoing health benefits for both infants and mothers [104,105]. As such, this review provides evidence of very low to moderate certainty that there is a role for DHIs as part of the health care provided to parents during the first 2000 days as a potentially scalable strategy for improving breastfeeding behaviors [106].

While the meta-analysis for EBF demonstrated considerable heterogeneity (I²=83%), this was only slightly reduced when we excluded high risk of bias studies (I²=74%). Tests for subgroup differences suggested that while in all cases the DHI favored improved EBF rates compared with control, there was a statistically significant subgroup effect based on DHI type (P=.02), parent type targeted (P=.0002), and study country income level (P=.0007). These statistically significant subgroup effects indicated that a treatment effect was greatest for interventions that combined mHealth or eHealth strategies with telehealth, DHIs that targeted only mothers and were delivered within upper-middle-income countries. However, in all cases, there remained substantial unexplained heterogeneity within subgroups, suggesting that the validity of the treatment effect estimate for each subgroup is uncertain, as individual trial results remain inconsistent. While GRADE assessment of the EBF evidence suggests a very low level of certainty (influenced by high levels of heterogeneity and study risk of bias), the calculated 95% prediction intervals suggest that future studies exploring the impact of DHIs on EBF could reasonably expect effects ranging between moderately reduced odds (OR 0.50) due to DHI exposure to a very large increased odds (OR 13.30), demonstrating a very high potential for benefit. In addition, our meta-analysis exploring the impact of DHIs on breastfeeding duration was statistically significant with no heterogeneity present (SMD 0.48; P=.0007; I²=0%). Additionally, calculated 95% prediction intervals indicated that future studies could expect DHI exposure to increase breastfeeding durations between 1 and 3 weeks (SMD 0.29-0.66) with moderate certainty in the evidence, adding to the overall findings that parent-directed DHIs are effective at improving breastfeeding outcomes.

A recent systematic review and network meta-analysis by Fan et al [107] indicated that text messaging interventions that covered antenatal and postnatal periods and delivered weekly were most effective in improving EBF rates. Meanwhile, our meta-analysis of DHIs more broadly indicated that the odds of EBF were not significantly modified based on whether the DHI was delivered only during the postnatal period (OR 3.51; P=.0003) or commenced during the prenatal period (OR 2.25; P=.0001). This result should however be interpreted with some caution, given that the number of studies included in the postnatal engagement subgroup is much lower (n=13) compared with the prenatal engagement subgroup (n=26). Thus, several other factors could be driving this result such as the broad differences between DHIs in terms of intensity, dosage, and the inclusion of peer or health professional contact. Given the diversity of included DHIs within this review, quantifying the impact of these factors was not possible within the current scope of this review. However, DHIs that were assessed at a lower risk of bias and produced clear improvements to EBF compared with control contained some consistent design elements. These elements related to the use of evidence and theory in guiding intervention content and design [56,64,76,87,91,97]; the inclusion of actionable or motivational content addressing common breastfeeding challenges [56,76,87,95,97]; and the DHI content was provided alongside another nondigital element such as a physical booklet, face-to-face education sessions, and/or existing health care model [56,64,87,95,97].

As birthing parents globally continue to face barriers to optimal breastfeeding behaviors [108,109], DHIs demonstrate huge potential to positively transform breastfeeding support, as they uniquely offer a highly scalable model to directly reach consumers with relatively high fidelity [26]. While this review, along with previously published literature, demonstrates efficacy of DHIs for improving breastfeeding outcomes [27-29,103], we have identified an ongoing need to explore and publish the safety (eg, unintended adverse events) and cost-effectiveness of such innovations. As published data for these key scalability outcomes remain scarce across the published literature [110], we recommend future DHI research prioritize collecting and publishing safety and cost-effectiveness data to support the case for countries to invest in digital health care policy and infrastructure. A review of reviews by Tomori et al [111] suggests that there is a need to build on well-established knowledge to scale up breastfeeding protection, promotion, and support programs, suggesting that the literature as a whole disproportionately focused on high-income and upper-middle-income settings. Tomori et al [111] also suggest that while DHIs offer a promising opportunity for improving breastfeeding outcomes, better quality research design is necessary to ascertain the most effective types of DHIs. Our EBF meta-analysis included 13 RCTs conducted in high-income countries, 15 from upper-middle, and 11 from low or lower-middle-income countries, suggesting a fair spread of RCT evidence informing our findings, including a broad variety of cultural and health system contexts. Further, the subgroup analysis by country income indicated statistically significant differences across subgroups (P=.0007), with the greatest effect noted for upper-middle-income countries (OR 4.35), followed by low or lower-middle-income (OR 2.73) and high-income (OR 1.45); yet, the positive impact of DHIs on EBF was statistically significant for all subgroups. However, given some of the inherent limitations of DHIs including a reliance on personal technology and local network access, varied engagement levels, and health or technology literacy disparities [112,113], concerns remain for breastfeeding inequities for socioeconomically vulnerable populations and those who belong to discriminated ethnic or racial groups [114,115]. As such, future studies should prioritize exploring the reach and impact of DHIs across population subgroups to ensure that population-level scale-up and the expected breastfeeding benefits would reach priority populations.

While the meta-analyses for breastfeeding duration indicated a consistent effect across individual DHI studies, with no heterogeneity (I²=0%), there was considerable unexplained heterogeneity present in the EBF meta-analysis. This heterogeneity is likely driven by the large number of included studies that carried differences in study risk of bias (eg, methodological heterogeneity) and differences in participants, types of DHIs, and outcome assessment (eg, clinical heterogeneity). For example, many studies included in the analysis (57%) were deemed to have a high risk of bias, diminishing the trustworthiness and generalizability of the EBF results, which has been reflected in our GRADE assessments. However, when we removed the high risk of bias studies, the meta-analysis indicated similar effect sizes and statistical significance.

Studies included within this review almost entirely relied on self-reported data to inform breastfeeding status despite previous research suggesting self-reported measures of EBF are likely overestimated due to parents’ misunderstanding of what constitutes EBF [116]. Further, varied definitions of what qualified EBF within studies were minimally provided and could also be influencing large variations. For included studies that reported on EBF and “any” breastfeeding across multiple time points, we included the data for the longest follow-up time point (ie, included 6 months data rather than 3-month data) within our meta-analyses. Given that there is an ongoing need to increase the proportion of infants EBF to 6 months and an increase in breastfeeding duration is most desirable [117], the inclusion of data to the longest point to follow-up aligns best with recommended breastfeeding behaviors. However, it is possible that this approach is underestimating the impact of some DHIs included within the review.

Higher breastfeeding self-efficacy scores are consistently linked with better breastfeeding outcomes [118]. Our meta-analysis suggests that parent-targeted DHIs have no impact on breastfeeding self-efficacy. However, given our meta-analysis only included RCTs that reported on a breastfeeding outcome aligned with our primary outcomes (as per systematic review inclusion criteria), this analysis is likely missing key studies to inform the impact of DHIs on breastfeeding self-efficacy; thus, the true impact of DHIs on this outcome could be underestimated and should be taken with caution. Additionally, consistent with previous systematic reviews of this literature [103], the use of “usual care control” groups likely contributed a major source of inconsistency between studies. This review included RCTs that targeted parents and primary caregivers generally, which is an extension from previous reviews that focused exclusively on mothers [103]. While the subgroup analysis of EBF indicated that DHIs targeting parents generally (ie, fathers or nonbirthing coparents) were not effective (OR 1.29), this subgroup included only 7 studies; thus, further research exploring this relationship could be beneficial, given fathers or nonbirthing coparents and caregivers are considered a key source of social support for mothers [119], and seems an emerging field of exploration.

Strengths of this systematic review and meta-analysis include the use of robust research methodology aligned with best practice Cochrane methodology [31] and include high-quality research (RCTs), providing a high level of evidence for the effect of DHIs on breastfeeding behaviors. Further, this systematic review includes studies from a variety of countries internationally, addressing a key limitation of previous systematic reviews conducted on this topic [27].

To the authors’ knowledge, this systematic review is the first to combine global RCT evidence spanning diverse cultural and health system contexts to comprehensively synthesize the impacts of parent-targeted DHIs (including eHealth, mHealth, and telehealth) on key breastfeeding outcomes. This review is innovative in that it has comprehensively combined data using best practice methodology to summarize evidence from a rapidly growing field, and to address an ongoing public health and policy priority. Given even minor improvements to EBF and breastfeeding duration are linked to significant societal and public health benefits [15], our findings indicate with very low to moderate certainty that parent-targeted DHIs may be an effective breastfeeding promotion strategy, specifically for EBF rates, and improved breastfeeding duration. It is important to highlight that the results of this review have been graded as moderate to very low certainty and should be interpreted cautiously, due to the presence of trial methodology biases, potential publication bias, and heterogeneity. Our findings indicate real-world implications for parent-targeted DHIs, suggesting that DHIs should cautiously be considered as a complementary strategy to support the usual health care provided to parents during the first 2000 days. Given the identified limitations in the quality and consistency of current evidence, future trials should seek to minimize possible biases by publishing comprehensive trial and data analysis protocols. Further, data pertaining to the potential adverse events, cost-effectiveness, and equity of DHIs are urgently needed to support justification for widescale implementation and policy investment.