To best address the study aims, we chose a cross-sectional research design using content analysis. Given that apps are updated with new features over time, a longitudinal design was not appropriate. Our methods were informed by previous studies of infant-feeding apps [20,23] and other health apps [24-26]. This study was exempt from Institutional Review Board approval.

To compile a convenience sample of breastfeeding apps, a graduate student in the US conducted a keyword search in the Apple App Store (iOS) and Google Play Store (Android) in fall 2018. A combination of keywords was used to search for English-language breastfeeding apps, including “breastfeeding” and “breastfeeding applications.” In January 2019, another graduate student created a sample in the same manner and cross-referenced it with the fall 2018 sample, increasing the sample size while also removing duplicates and dead links; the sample was finalized in February 2019. All relevant apps were included regardless of their cost. All apps were free except for 9; these 9 paid apps were downloaded for a combined cost of US $31.92 ($18.95 for 5 iPhone apps and $12.97 for 4 Android apps). A total of 40 iPhone and 42 Android apps were included in the final sample (N=82) of which 80 were free to users; only 2 paid iPhone apps remained in the final sample (Figure 1). The final sample is comparable to those of previous infant-feeding app studies, which included 41 to 77 apps [18,20,21].

To analyze the apps, a coding schema was created a priori based on existing studies of apps [20,24,27] and 2 breastfeeding textbooks [28,29]. The schema contained 87 distinct app characteristics and features. We defined features according to Sidhu et al [21] as any “opportunity for user interaction with the app (e.g., a button).”

Descriptive characteristics were derived from the app’s download page and included the name of the app, website link, download date, version number, date of last update, developer or seller name and affiliation (ie, commercial, government, nongovernment organization, university, unknown, or other), whether and which experts or end users were involved in the app development process, user rating (ie, number of stars out of 5), number of user reviews, app category (ie, medical, lifestyle, health and fitness, parenting, or other), language options, cost of basic and premium app versions, and age rating (not unlike a movie rating, each platform recommends the minimum maturity level of app content for end users by age, ie, >0, >4, >12, or >17).

Features were observed by navigating the downloaded app. Tracking features monitored breastfeeding, bottle feeding, solid foods, pumping and human milk expression, diapering, bathing, sleeping, infant growth and development, medication and vitamin use, vaccinations, temperature, illnesses, and well-child visits. Nontracking features included the ability to add notes, information, pictures, or videos; connect to a breast pump; print or export data; sync data with another program or device; use the app for more than one child or for multiple caregivers to use the app; customize features; receive static information (ie, articles, guidance, tips, checklists, product recommendations, frequently asked questions, pregnancy information, maps, graphs, or charts); share data with others (ie, other caregivers, health care providers, or social media); and interact with peers, lactation professionals, or others (Table 1).

All apps were downloaded to a donated iPhone (iPhone 5S, iOS version 12.1.4; Apple Inc) or Android smartphone (Samsung Galaxy Note 2, Android version 4.4.2, or Samsung Galaxy S8, Android version 7.0; Samsung Electronics Co Ltd). Coders primarily used the shared iPhone and Android phones, aside from 1 student who used a personal iPhone to expedite coding. Coding of the apps began in March 2019 and was completed in December 2019. Two graduate students and 1 undergraduate honors student (with no prior involvement in the study) from different academic departments (including food and nutrition, law and governance, and political science and law) coded the apps.

Interrater reliability (IRR) was measured among the 3 coders with 3 possible outcomes. Agreement between all 3 coders was labeled as “complete agreement.” Agreement between 2 of 3 was considered “partial agreement.” When all 3 coders disagreed, we deemed this “no agreement.” In 7 apps, only 2 coders completed the coding; thus, the IRR was determined as “complete agreement” or “no agreement.” This might have occurred because one of the students used their personal phone or used a phone that was incompatible with a particular app version. The authors reviewed coder agreement on variables with partial or no agreement to determine the final coding decision. For continuous variables (eg, the number of languages, user ratings out of 5 stars, and number of user ratings) we used the most recent version of the app.

To address aim 1—characteristics of breastfeeding apps—descriptive statistics were used to characterize the sample and are reported as frequencies and percentages. Apps rated ≥4.5 stars were defined as higher-rated apps, while those rated <4.5 stars were considered lower-rated apps. Prior studies have used a cutoff of ≥4 stars [18,20]; however, the present sample had a skewed rating distribution, whereby only 16% (13 of 82) of apps were rated under 4 stars. Therefore, the 4.5-star cutoff was chosen to maximize variability in both groups. Nine summative indices were created by grouping like features by topic (Table 1). For aim 2—comparison of higher- and lower-rated apps—we used the Shapiro-Wilk test to assess the normality of the data and found that the indices were not normally distributed. We conducted Mann-Whitney U tests, which are appropriate for nonnormally distributed independent groups, to assess whether higher- and lower-rated apps differed by index.

To address aim 3—predictive relationships between user ratings and indices—we determined the Spearman rank correlation between the indices, user star ratings, and whether the app was higher rated or lower rated. All indices that were significantly correlated with the user star ratings were included in the linear regression model, except for bottle feeding, which was highly correlated (r=0.693) with the breastfeeding index. The same indices were entered into a binary logistic regression model to examine their ability to predict whether an app was higher or lower rated. Logistic regression results are reported as odds ratios (ORs) with 95% CIs. For all statistical tests, significance was defined at P<.05.

The sample was composed of 82 breastfeeding apps, including 40 iPhone and 42 Android apps (Table 2). On average, users rated breastfeeding apps 4.4 of 5 stars. Of the 82 apps reviewed, two-thirds (54) were higher rated and one-third (28) were lower rated. The number of user ratings per app ranged from 4 to 81,800.

Mann-Whitney U tests were performed (Table 3) to determine differences between higher- and lower-rated apps. All indices were significant, and the mean ranks for all indices except the informatics, informational, and interactivity indices were greater among higher-rated apps than lower-rated apps. The breastfeeding and solid-food indices yielded the most notable differences in median scores between higher- and lower-rated apps.

Table 4 illustrates the Spearman rank correlations between user star ratings, whether an app was higher versus lower rated, and the indices. The correlation between user star ratings and the solid-food index was positive and strong, while the correlations for breastfeeding, bottle-feeding, and technical indices were positive and moderate. The correlation between an app being higher rated and the solid-food index was positive and strong, while the correlations with the breastfeeding index were positive and moderate. Finally, the correlations between an app being higher rated and the bottle-feeding and technical indices were positive and weak.

A linear regression analysis was performed to determine whether breastfeeding, solid-food, and technical features predicted user star ratings (Table 5). The independent variables explained 17% of user star ratings (adjusted R²=0.172). The breastfeeding and solid-food indices were significant. For each additional breastfeeding feature, we can expect to see a 27% (β=.265, P=.047) increase in the user star rating, while each additional solid-food feature increases the user star rating by 35% (β=.354, P=.009).

A binary logistic regression analysis was performed to determine whether tracking features or nontracking features predicted higher user star ratings (Table 6). In the unadjusted bivariate analysis, there was a significant association between the breastfeeding, bottle-feeding, solid-food, and technical indices and the dependent variable. In addition, the odds of an app receiving a higher rating increased by 28% (OR 1.284, 95% CI 1.064-1.550) for each additional breastfeeding feature. Similarly, the unadjusted odds of an app receiving a higher rating increased by 68% for each additional bottle feeding (OR 1.683, 95% CI 1.112-2.548) and solid-food (OR 1.685, 95% CI 1.236-2.297) feature. The technical index also increased the odds that an app was higher rated. In the adjusted model, only the solid-food index remained significant. The odds of an app receiving a higher user star rating increased by 58% (OR 1.579, 95% CI 1.074-2.321) for each additional solid-food feature.

Our study builds on previous research of breastfeeding apps while expanding our understanding of what these apps offer by evaluating their features. Our sample is slightly larger than that of Mieso et al [18] and includes a greater percentage of free apps than earlier studies [18,20,21]. Similar to Schindler-Ruwisch et al [20], the sample draws upon a range of app categories, including medical, health and fitness, and parenting. Our cross-sectional review of apps occurred within a specified timeframe, akin to earlier studies [20,21].

Many characteristics of our sample reflect earlier studies of breastfeeding apps. For example, breastfeeding apps tend to be highly rated. Both Mieso et al [18] and Schindler-Ruwisch et al [20] found that nearly 70% of apps received user ratings >4 stars, and Mieso et al [18] showed that the average rating for breastfeeding apps was 4.3 of 5 stars. This is consistent with our findings. Similar to Mieso et al [18], the number of user reviews in our sample displayed a wide range.

Unsurprisingly, higher-rated apps offered more tracking features on all indices. In their qualitative analysis of maternal and infant health app user reviews, Biviji et al [30] found that across positive reviews, many users mentioned tracking features, including feeding, pumping, diapering, and sleep—akin to one-stop shopping. Conversely, there were complaints about apps with limited data-tracking abilities [30]. According to Mendiola et al [31], factors that predicted user ratings of health apps include usability, data export, and tracking. While the tracking component was negatively associated with user ratings, it was positively correlated with export and usability, both of which were positively associated with user ratings [31]. An alternative explanation as to why more features might appear in higher-rated apps is the release of new app versions that include new or updated features. Future studies should consider how tracking features correspond to other usability features and critically analyze the tracking features to determine their appropriateness to support infant-feeding goals. While informatics, informational, and interactivity features were not correlated with user star ratings, lower-rated apps had higher scores for these indices. Though few apps appeared to have these features, future studies might consider investigating their utility, since the study by Biviji et al [30] suggests that users desire these features.

In the regression models, the breastfeeding index predicted user star ratings; however, it did not predict whether an app was higher or lower rated. The former finding is to be expected, since breastfeeding tracking is the primary purpose of the apps. This is supported by Sidhu et al [21], who found that apps often had features that assisted with human milk tracking. However, we are unable to explain the latter finding, though it may be related to how we defined higher-rated apps.

Across both regression models, the solid-food index was significant. Solid food–tracking features allow users to continue with a familiar app that contains other tracking data (such as human milk or human milk–substitute consumption, diaper changing, or vaccinations) by carefully monitoring the introduction of new foods, which typically occurs on a weekly basis. This prolongs the usefulness of an app beyond a limited timeframe, again tapping into one-stop shopping [30]. Biviji et al [30] reported that positive app reviews emphasized tracking, highlighting feeding in particular. The authors demonstrated how users provided additional feedback on exporting data, additional tracking options, and data visualization, which might be incorporated into updated app versions [30]. Alternatively, the solid-food index might be a proxy for a feature not included in our study. Since our model only explains a small portion of the variance, we recommend an overall assessment of an app’s interface. For example, how seamless are the features? What are the advantages offered by one app over another? We also recommend a qualitative study of breastfeeding app users to gain greater insight into the reasons behind app adoption and features utilized.

While this study provides an overview of breastfeeding apps, there are several limitations. First, this was a convenience sample gathered between November 2018 and December 2019. We conducted a manual search with keywords, which may have resulted in missing some apps. Second, since new apps are frequently introduced to the market, this research only provides a snapshot in time; however, with a total of 82 apps, it still offers a comprehensive overview. Third, this research is limited to English-language apps in the US. Future studies should consider apps in other languages and in countries with higher rates of breastfeeding. Fourth, this study does not examine the apps’ clinical or scientific merits, but instead assesses features. Breastfeeding apps might contain content that is contrary to medical advice, and apps might not conform to national guidelines on infant feeding [19,23]; nevertheless, this was beyond the scope of this study.

This study of breastfeeding apps demonstrates that user ratings are partially driven by tracking features, specifically those related to breastfeeding and solid foods. Nontracking features appear to be less important with regard to how users rate apps, though why this is the case remains unclear. Researchers should consider investigating this in the future. More importantly, the proliferation of mHealth offers opportunities for parents and caregivers to track behaviors associated with infant feeding and other health metrics in a dynamic, detailed, and comprehensive manner. In this way, breastfeeding apps have the potential to promote and support breastfeeding among users.