The eligibility criteria for participation in the intervention group (GH) included being pregnant (30+ weeks gestation) or parent or main carer of an infant younger than 3 months, ownership of any type of mobile phone, ability to speak and read English, age 18 years or older, and residing in Australia. Participants were recruited using 3 methods: via their primary care providers in socioeconomically disadvantaged communities in 2 Australian states; face-to-face by researchers in first-time parent groups; or Web-based advertising. A concurrent nonrandomized comparison group (Baby’s First Food, BFF) was recruited via online forums, social networking sites, and blogs and received usual care. The eligibility criteria for participation were the same as the intervention arm with the exception that participants were not required to own a mobile phone. Enrollment to both groups of the study involved the completion of a Web-based screening form, a consent form, and a baseline survey. Further details of the recruitment process and outcomes have been published elsewhere [34].

In brief, the GH program aimed to encourage parents to engage in infant feeding practices that promote healthy rather than excess weight gain, with a focus on socioeconomically disadvantaged parents. The aims of the program were as follows:

The main delivery media for the program were an app and website, which provided parents with evidence-based article and videos containing practical advice and strategies consistent with national guidelines on infant feeding from birth until 9 months of age. The development of the program was guided by the Behaviour Change Wheel and the Capability, Opportunity and Motivation model of behavior change [35]. For each program aim (target behavior), key determinants were identified using prior formative work [36,37] and literature and mapped to intervention strategies. Participants received 3 personalized push notifications or text messages (short message service, SMS) per week targeting specific intervention strategies and behavior change techniques as detailed in our protocol paper [33]. Push notifications were also tailored to each infant’s age and stage of development as well as their feeding mode (breast, formula, or mixed feeding), directing them to relevant content in the app. A weekly email was also sent that included the 3 messages for the week with links to the website. This was introduced part way through the intervention in response to the low number of push notifications being opened. Participants were also invited to join a Facebook group where one feeding message per week was posted by a moderator, and participants were encouraged to discuss practical experiences around infant feeding.

Data were collected at 3 time points via a Web-based survey: when infants were less than 3 months old (T1), when the infant was 6 months of age (T2), and when the infant was 9 months of age (T3). To compensate participants for the time involved in completing surveys, GH participants received a gift voucher worth Aus $20 per survey completed, and BFF participants received Aus $40 for the completion of 2 or more surveys. Nonresponders to the survey were sent 3 email reminders, 1 week apart. We also collected data from analytics within the app.

Basic descriptive analysis and cross-tabulations were calculated. Baseline characteristic comparisons were made using chi-square test and t test as appropriate to the variable type. Survival analysis was used to assess the difference in timing of feeding practices, including duration of any breastfeeding, timing of introduction of infant formula, and timing of solids introduction between the 2 groups—GH and BFF. Kaplan-Meier survival curves were used to assess the mean and median of breastfeeding duration and the timing of introduction of solids. Differences between the groups were assessed using Breslow test. Those cases where the child was reported as still being breastfed at T3 were classed as censored observations in the breastfeeding duration analysis. Cox proportional hazards regression models in the analysis of breastfeeding duration and time to solids introduction were used to account for covariates such as child’s gender, whether first born, dummy use, maternal smoking status, mother’s country of birth, parental education and work status, household income, and maternal age and prepregnancy BMI. Mother’s self-rated health and possession of a health card were considered but due to high correlation with other covariates were excluded from the final analysis. Evaluation of the log-minus-log survival curves provided no evidence that the assumptions of proportional hazards were not met.

Multivariate logistic regression was used in the analysis of differences between GH and BFF for binary outcomes, including proportion exclusively breastfeeding; formula feeding outcomes; and proportion who had introduced solids before 4 months, after 6 months, and at 4, 4.5, 5, 5.5, and 6 months.

Differences between GH and BFF group continuous variables, which included child BMI z-score, weight, and length at time points between birth and T3, were assessed using mixed effect polynomial regression models with an unstructured covariance structure. A random intercept and a random slope for age to allow individual growth rates were fitted. Quadratic and cubic age variables to model nonlinear growth were included. Covariates included in the model were child’s gender, whether first born, dummy use, maternal smoking status, mother’s country of birth, parental education and work status, household income, and maternal age and prepregnancy BMI. All statistical analysis was performed using IBM Corporation SPSS version 24 [41].

A total of 909 subjects commenced enrollment into the study; however, 264 were ineligible predominately because they did not complete the baseline survey or their baby was older than 15 weeks or born prematurely (Figure 1). The final sample included 645 carer/child dyads at baseline (GH: 301; BFF: 344). Retention to the study was high at T2 (84.7%) and T3 (80.3%; Figure 1).

At enrollment, the mean age of the children in the GH group was significantly younger than the BFF group (7.0 weeks compared with 7.9 weeks). Furthermore, the GH group compared with the BFF group contained a significantly higher proportion of first-born children, with GH mothers being younger (30.4 years compared with 31.2 years), a lower proportion of GH mothers being Australian born (84.1% vs 90.1%), and a lower proportion in the highest household income category (29.4% vs 36.8%). GH also contained a lower proportion of breastfeeding mothers and higher proportion of formula and mixed feeding mothers at baseline when compared with BFF (P<.001). Other characteristics considered were not found to be significantly different at baseline between the 2 groups. Details of the baseline characteristics of study participants are provided in Table 1.

Of the 301 participants, 260 (86.4%) opted to access the program via the app, and 41 (13.6%) via the website/SMS. App participants were provided with a code to enable them to download the app from either Google Play or the App Store, and 74.8% (225/301) of participants downloaded the app. More than half of the sample used iPhones (71.6%), with 28.4% using Android phones and 11.9% website and SMS. Of app users, Android phone usage was higher among non-university-educated participants (31.7%) compared with university-educated participants (24.5%). App users were sent 3 push notifications each week, and on average, 11 push notifications were opened over the time of the study (8.0% of all notifications). App usage declined over time from 92.0% using the app at least once on enrollment to 38.2% at study completion when infants were aged 8 to 9 months of age (Figure 2), with a similar decline in the mean number of sessions using the app across the duration of the study (Figure 3).

Overall, participants reported high levels of satisfaction with the program, with 88.1% agreeing that they liked the program and 84.6% informing that they would recommend it to a friend (Table 2). Most parents (86.1%) reported that the app provided trustworthy information and was easy to understand (91.0%) and use (78.3%), with less than 5% of parents expressing concern about data usage when using the app. However, just over a quarter of parents reported that the app failed to work at times. Nearly 20% of participants who completed T3 reported disabling push notifications on their phone. For those receiving push notifications, a majority found them helpful, well suited to their baby’s age and stage of development, and appropriate in terms of number and timing of messages. However, just over a third reported that the messages disappeared before they read them, and nearly 40% were unsure how to retrieve push notifications once they had disappeared from the screen.

This is the first study, to our knowledge, to report on the feasibility and effectiveness of an mHealth intervention for prevention of obesity in infancy. Our findings support the use of mHealth as a feasible and acceptable mode of delivery of an intervention targeting parents’ infant feeding behaviors due to participants’ high levels of reported satisfaction and retention to the program, despite some technical difficulties using the app and decline in engagement over time. We were, however, unable to demonstrate any impact of the intervention on the target behaviors and outcomes of breastfeeding duration or exclusivity, timing of introduction of solids, or infant growth trajectories. The findings suggest that the GH intervention may have positively impacted on formula preparation practices, although this was not significant in fully adjusted models and requires confirmation in an adequately powered randomized controlled trial.

The findings of this study support mHealth as an acceptable mode of delivery for obesity prevention interventions in infancy, particularly those targeting infant feeding, with high rates of recruitment [42] and a retention rate of 80% at 9 months follow-up. It is important to acknowledge that the high retention rates might reflect, in part, the payment offered for survey completion and the use of 3 reminders. Acceptability of the program is supported by the high rates of reported user satisfaction with the program. This is consistent with findings from our qualitative follow-up interviews with parents [43] where they reported engagement with the program was promoted by the credibility of the program source, the user-friendly interface, and tailoring of content and push notifications to the baby’s age and key transition points. Our findings are congruent with existing research, which suggests that parents are increasingly relying on online source of information for infant feeding and care [23-27].

However, a number of factors may have reduced engagement with the program. Over a quarter of participants reported that the app failed to work at times. Technical problems did arise in the study, including operating system updates for both iOS and Android systems disabling the app for a short period of time. Parents who changed mobile phones during the study were required to contact the research team to obtain another code to access the app, which may have further reduced app usage. Furthermore, although push notifications were perceived to be relevant and timely, nearly 40% of participants were unsure how to retrieve push notifications once they had disappeared from the screen. This might explain the relatively low proportion (8%) of push notifications opened. Given that push notifications were the primary mechanism to drive parents to engage with the app content, this was very likely to have limited the dose of the intervention received and its subsequent impact on infant feeding behaviors and outcomes. This is further supported by app analytics data, which indicate that the use of the app did indeed decline over the duration of the study.

Understanding factors influencing engagement with mHealth programs and how these can be maximized over time is critical to program effectiveness. Our quantitative analysis of app usage [44] revealed higher engagement with the program among those recruited by their health practitioner, those who registered when their infant was younger, those who were first-time mothers, and those using both the app and website (via email links) compared with those using the app alone. This suggests that, to maximize engagement and potential impact, consideration should be given in the future to focus on recruiting first-time parents, with the help of health practitioners during the early postnatal or antenatal period. Future iterations of the program should include design features to improve access to push notifications and to maximize engagement over time using multiple methods (eg, email and push notifications) as well as the inclusion of more interactive features such as a forum and other tools to promote ongoing engagement. Technical issues also need to be addressed in a timely manner, including accommodating any operating system updates to ensure the program is functional at all times.

In addition to intervention dose, the lack of intervention effect may also be because of the sample size of this feasibility study, which was not powered to detect differences between groups but rather inform sample size calculations for a subsequent larger randomized controlled trial. The study was limited by the significant difference in a number of baseline characteristics between the study groups. Although baseline differences were controlled for in the statistical analysis, this reduces the power to detect differences in key outcomes between the groups.

The timing of the intervention may have contributed to the program’s limited impact on outcomes, particularly for breastfeeding. The average age of infants at the time of enrollment was 7 to 8 weeks, suggesting that the intervention missed the critical period for breastfeeding support, with national data indicating that 40% of mothers introduce formula by 1 month of age [14]. Research has also indicated that plans about whether a mother will breastfeed and for how long are made antenatally [45], and this was consistent with our qualitative findings [43] where mothers reported that plans for feeding their infant were made before enrolling in the program. This highlights the importance of commencing the program before birth and providing very early postnatal breastfeeding support to influence breastfeeding outcomes.

The GH program specifically targeted socioeconomically disadvantaged parents due to the socioeconomic disparities in obesity risk emerging in early infancy [17]. Our findings suggest that we had some limited success in reaching these parents, with just over half of mothers (51.6%) and nearly 70% of fathers not having a university-level education, with education commonly used as a proxy for socioeconomic position [46]. This is similar to the national average [47] but higher than that reported in other group-based obesity prevention trials in infancy. For example, the proportion of mothers without a university education was 46% in the InFANT trial [48] and 42% in the NOURISH trial [49]. However, a home-visiting trial in disadvantaged communities managed to recruit three-quarters of participants without a university education [50]. Surprisingly, we found no difference in the education levels of mothers when recruited by primary health care practitioners in disadvantaged communities and those recruited via social media [42]. It may be that primary health care practitioners were more selective in offering the program to less vulnerable parents. Our findings indicate the importance of catering for Android phone users if mHealth programs target socioeconomically disadvantaged parents, with higher Android phone usage among those without a university education compared with university-educated mothers in our sample. Given the greater need in socioeconomically disadvantaged parents, further research is required to ascertain how best to engage these parents in obesity prevention interventions and indeed whether mHealth programs provide a useful mode of delivery.

The use of mHealth for obesity prevention in early childhood is a rapidly growing field, with a number of trials underway [51-54]. However, we are unaware of any other studies reporting the outcomes of a mobile phone app targeting parents’ healthy infant feeding practices. A recent meta-analysis of 16 studies in developed countries has shown e-technologies (including SMS, Web, and interactive computer agent) to be effective in improving rates of breastfeeding initiation, duration, and exclusivity [55]. This review however did not include any studies utilizing a mobile phone app. Only one other published study [56] has reported the effectiveness of an mHealth program targeting infant feeding. That study [56], by Jiang et al in Shanghai, China, found that a weekly SMS from third trimester of pregnancy to 12 months postpartum resulted in a significantly higher rate of exclusive breastfeeding at 6 months and a significantly lower rate of the introduction of solid foods before 4 months. However, the intervention had no effect on other infant feeding practices, including taking a bottle to bed, drinking from a cup, or using food as a reward.

The key strength of this study was that the intervention was informed by behavior change theory and extensive formative work and had high rates of retention. The use of a quasi-experimental study design was a limitation with a number of baseline differences between the intervention and the comparison groups; however, these differences were controlled for in the statistical analysis. There was also a likely selection bias in that the mothers who took part in both groups may have been more interested and motivated to achieve desirable infant feeding practices by virtue of their interest in this research despite having similar levels of education as national average [47]. This is supported by the high rates of exclusive breastfeeding at baseline (84% in BFF and 82% in GH compared with national average of 48% in infants less than 3 months of age [14]) and low proportion introducing solids early (13% at 4 months in both groups compared with national average of 35% [14]). This may have limited the ability to detect intervention effect. A randomized recruitment strategy with a more representative sample of mothers might offer more potential to show improvements in infant feeding practices. Measurement of infant feeding behaviors based on self-report in both groups was no doubt subject to social desirability bias. Parental report of weight and length may be subject to transcription errors and would be more accurately collected by investigators.

Our study design was feasible in that there was an excellent retention rate over time for participants who completed the enrollment survey, and the results provide a useful estimate on which to base a sample size calculation for a larger study. An mHealth intervention using a smartphone app to promote healthy infant feeding behaviors is a feasible and acceptable mode for delivering obesity prevention intervention to parents but further work is required to sustained engagement and use over time. The limited impact of the program on key measureable infant feeding outcomes may reflect that some participants received a low intervention dose because of unforeseen technical problems, the timing of the program, participant selection bias, and/or limitations in the study design. It is recommended that mHealth programs targeting infant feeding commence antenatally and future iterations of the program have contingencies in place to address technical issues in a timely manner and design features to maximize engagement over time. Future research using larger randomized controlled trial designs are required to determine the effectiveness of mHealth programs for obesity prevention in infancy.