All randomized controlled trials (RCTs) and non-randomized studies including controlled before-after studies, interrupted time-series studies, and prospective comparative cohort studies were considered for inclusion. Case-control studies and cross-sectional studies were excluded due to their retrospective design. Crossover trials were also excluded as they are considered most suitable for temporary effects and chronic conditions [15]. Women at any stage of pregnancy or labor were considered for participation.

Studies assessing the effects of mobile app-based interventions designed to influence maternal knowledge or behavior during pregnancy were considered for inclusion if they provided general information for pregnant women or focused on a specific maternal risk factor or perinatal outcome. There was no restriction on who sponsored the intervention or type of setting.

Studies were excluded if the intervention (1) did not utilize a mobile app, (2) used a mobile phone solely for telephone conversations or SMS (short message service) text messaging, (3) did not report on maternal or infant health outcomes, (4) did not target pregnant women (focus on clinicians, partners), and (5) focused on physical effects of mobile phone usage (such as radiation) during pregnancy.

The following comparisons were planned:

The primary outcome measure was a change in maternal behaviors (as defined by trial authors), by intervention goals. Secondary outcomes addressed maternal and infant health outcomes, maternal experience with the intervention, and health service utilization measures (Table 1).

Systematic searches were performed using seven electronic bibliographic databases: PubMed, Embase, The Cochrane Library, CINAHL, World Health Organization Global Health Library, POPLINE, and CABI Global Health. Handsearching of journals and conference proceedings from the reference lists of retrieved studies were also conducted. No language or date restrictions were applied. Abstracts and full-length articles were obtained for each citation, where available.

The specific search strategies were developed by the primary author (LMD) and an experienced clinical research librarian, with input from all authors. Electronic searches using subject headings and all fields for keywords were conducted to avoid missing non-indexed concepts. Search terms and returns by the database and handsearching are presented in Multimedia Appendix 1. Articles published in non-peer reviewed publications were excluded, as per the review protocol. Remaining citations and abstracts were uploaded to the Web-based software platform Covidence [16]. Throughout the review process, authors were not blinded to journal titles, study authors or institutions. If it was unclear whether studies met inclusion criteria, study authors were contacted up to two times by email to request further information.

Abstracts of articles retrieved through the search strategy were independently screened by 2 review authors to determine if inclusion and exclusion criteria were met. A third author addressed any concerns about inclusion. If necessary, additional information was sought from study authors to resolve eligibility queries. For selected studies, full-text articles were obtained and read by 2 authors to confirm that they met inclusion criteria.

Studies were assessed for quality by 2 reviewers independently (LMD, VF), according to the guidelines in the Cochrane Handbook for Systematic Reviews of Interventions [17], by 6 domains: (1) selection bias, (2) performance bias, (3) detection bias, (4) attrition bias, (5) reporting bias, and (6) any other possible bias. All included studies were assessed on the risk of bias, likely magnitude, and potential impact on findings.

Once the studies were selected, using standardized forms, 2 reviewers (LMD, VF) independently extracted data including study objective, study design, inclusion and exclusion criteria, data sources, study period, methodology, sample size, intervention details and effects, and outcomes. Due to the complexity of the interventions found, a post-hoc decision was taken to collect data on the interventions based on the Template for Intervention Description and Replication (TIDieR) checklist [18,19].

As described in the review protocol [20], we set out to synthesize data and present measures of treatment effects including summary risk ratios for dichotomous outcomes and mean difference for continuous data and subgroup analysis. Due to the considerable heterogeneity of participant characteristics, intervention features, and reported outcomes, it was decided that meta-analysis could not be performed, and results were summarized in a narrative synthesis.

The search strategy for this review has been consolidated into a Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) diagram (Figure 1), summarizing inclusion and exclusion of studies [21]. For database screening, one author (LMD) searched the databases on 15-16 February 2017, with a yield of 5,089 articles. After initial screening to remove duplicates and articles from non-peer reviewed journals, the titles and abstracts of 2,241 articles were reviewed by 2 authors independently (among LMD, VF, DH, PFM, and FMB). Title and abstract screening of 1,091 additional articles identified through handsearching was performed by 2 reviewers (LMD and VF); however, no additional studies were identified. Full-text screening of 69 articles was performed by 2 review authors (among LMD, VF, PFM, and DH). At all stages, disagreements between authors were resolved by consultation with a third reviewer. Reasons were recorded for excluding studies (Multimedia Appendix 2). Several articles had multiple reasons for exclusion, although each was allocated to a single category. A total of 4 trials met the inclusion criteria.

The characteristics of included studies are presented in Table 2. Though not specified as a requirement for inclusion, all studies that met inclusion criteria used RCT designs and involved pregnant women in high-income countries. Three trials were based in hospital settings, comparing women using mobile apps with standard antenatal care. One community-based trial gave all participants a device to promote physical activity; participants in the intervention arm were also given a mobile app.

Objective assessments and validated data collection tools were employed in all included studies. Studies performed generally well regarding the risk of bias in random sequence generation; 3 studies were low risk and 1 study was unclear, as it was not described. High risk of performance bias was shared across all studies. Due to the nature of mobile app interventions, blinding participants is difficult, and did not occur in any of the included studies. Blinding health care providers can also be difficult and occurred in only 1 study. Most studies had a low risk of attrition bias, with low rates of withdrawal, drop-out or loss to follow-up. Reporting biases were also low, with all studies reporting results for their respective primary outcomes. The overall risk assessment is presented in Multimedia Appendix 3.

The primary outcome of interest was a change in maternal behaviors (as defined by trial authors), by intervention goals. All studies reported some type of behavior change and better results for the intervention group than controls for their respective primary outcomes (Table 5). Inventories and data collection tools used in the included studies are listed in Multimedia Appendix 5. The Ainscough et al study [22] concluded that a significantly higher proportion of the intervention group had transitioned to a “maintenance stage” of healthy lifestyle behaviors by 28 weeks’ gestation, compared to the control group (52.8% versus 32.7%, P=.004). The primary outcome measure for the Choi et al study [23] of physical activity was weekly mean steps, and intervention participants had a greater increase in daily steps at 12 weeks with 1096 (SD 1898) steps, compared with 259 (SD 1604) steps among control participants (P=.13). The change between groups reported across the 12-week study period was not statistically significant (P=.38). The Ledford et al study [24] found that by 32 weeks’ gestation, participants using a mobile app recorded information more frequently than the control group, and that they had developed greater “patient activation” than the control group (F [1127]=4.99, P ≥.05, n²=.04, marginal mean of 79.88 versus 74.81). The Zairina et al study [25] reported that the intervention group had a higher proportion of participants with well-controlled asthma than the control group (82% versus 58%, P=.03) at 6 months from baseline.

Of the 4 studies in this review, 2 studies [23,25] report maternal secondary outcomes relevant to this review, as further detailed in Multimedia Appendix 6. One study of asthma control reports a clinically significant improvement in the asthma-related quality of life among the intervention group compared with usual care at 6 months from baseline, but the mean change was not statistically significant [25]. Another study of physical activity during pregnancy reported reduced barriers such as lack of energy, time, and willpower among the intervention group, and decreased severity of pregnancy symptoms [23]. No studies reported data on major adverse maternal outcome, maternal knowledge about the targeted health topic, maternal evaluation of the intervention, or successful initiation of breastfeeding. Furthermore, none of the trials report statistically significant differences in neonatal outcomes between intervention and control groups.