Study procedures followed Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. We conducted a comprehensive literature search in online databases including PsycInfo, PsycARTICLES, Sociological Abstracts, Social Services Abstracts, Medline, ERIC, CINAHL Complete, and PubMed. We searched relevant studies published in and before June 2018. Advanced searches in titles, keywords, and abstracts were performed using the combinations of three groups of terms: (1) mobile technology and social media, including smartphone, mobile phone, social networking, Facebook, Twitter, WhatsApp, WeChat, and virtual reality; (2) pregnancy status, including pregnancy, pregnant, gestation, postnatal, and postpartum; (3) pregnancy care, including intervention, program, treatment, prevention, education, and therapy. In addition to the electronic database search, we hand-searched the reference lists of retrieved studies and relevant reviews, as well as grey literature including conference abstracts and dissertations.

The literature search aimed at identifying original evaluation studies on mHealth apps and social media interventions for pregnancy care. Eligible studies should (1) focus on interventions providing pregnancy care, including prenatal and postpartum health care for expectant mothers; (2) use advanced technology, such as mobile apps for social media or health care; (3) aim at health outcomes such healthy pregnancy and maternal well-being; (4) use experimental or quasi-experimental design; (5) report enough data to calculate the effect size; and (6) be published in English or Chinese.

Studies were not included if they (1) examined the use of mobile technology among health workers, (2) used a qualitative evaluation method only, or (3) used a traditional method such as providing short message services (texts) or sending emails.

In the first step, we designed a standardized form to code study characteristics. The study publication information (author, contacts, publication year, and country), methodological characteristics (study design, sample size, and the use of clinical sample or community sample), intervention details (aim, content, device, mHealth or social media apps, duration, attrition rate, and service provider), and participant profiles (age, pregnancy status, health status, and socioeconomic status) were recorded using this form. In the second step, we coded study outcomes and extracted data (eg, mean, standard deviation, P values, sample sizes) for effect size calculation. The outcomes include health outcomes of pregnant or postpartum women such as pregnancy weight control, asthma control, health knowledge, and stress and depression management. Two authors performed data extraction separately, and disagreements were resolved by consensus.

To obtain a valid estimate of intervention effectiveness and reduce the risk of bias in the meta-analysis, we used a checklist to assess the methodological quality of the included studies. The checklist (see Multimedia Appendix 1) is composed of eight items, measuring study design, participant eligibility criteria, sample size calculation, randomization process, intervention details, participant profiles, primary outcomes, and statistical methods. Each item is allocated 1 point; therefore, the highest score is 8 if all criteria are met, and scores of 5 and above are regarded as satisfactory. Two researchers in the research team evaluated the studies independently. To measure rater agreement, the Cohen kappa coefficient was used. The level of agreement was high between the two raters. Disagreements were resolved by discussion with the first author through a consensus-building process.

First, we provided a complete summary of the studies by tabulating publication information, methodology, intervention, and participant characteristics. Second, we calculated the effect size of each study using the Cohen d statistic. The Cohen d was calculated using the formula seen in Figure 1, in which the difference between two means is divided by a standard deviation for the two groups [25]. If a study reported multiple outcomes, the mean effect size of these outcomes was used. If there were studies based on the same intervention program, they were merged into one study and we calculated the mean effect size for these studies. The overall pooled effect size of the included studies was calculated based on a random effects model because of the different features of the interventions. The Q statistic and I² were used to measure the variation in study outcomes between different studies. In addition, we used the Q statistic to test the effect of moderator variables, which may be related to program effectiveness. To examine whether publication bias occurred in our meta-analysis, we constructed a funnel plot. A symmetric inverted funnel plot indicates an absence of publication bias with high probability. Statistical analyses were performed using the Comprehensive Meta-Analysis (CMA) 3.0 program (Biostat Inc).

Table 1 shows the effect size of each study pooled by different outcomes and time points. The results are displayed in a forest plot as shown in Figure 3. With the exception of the study by Olson et al [38], all of the interventions reported positive effects with an overall random effect size of 0.74 (P<.001, Q=146.45, I² =90.44). The overall effect size is considered medium to large, according to the Cohen criteria for effect size interpretation [39]. However, the large Q statistic indicated the wide variance in the effect sizes of different interventions, and it was estimated that 90.44% of the variance was due to heterogeneity. One reason for the large amount of heterogeneity is that the included studies aimed at different health outcomes. When we examined the effect sizes of different outcomes, the Q value decreased, showing that the variation is smaller for different specific outcomes.

As shown in Table 2, twelve studies aimed to improve the physical health outcomes of pregnant or postpartum women, including weight management (d=0.45, P=.003), GDM control (d=0.41, P=.03), and asthma control (d=3.43, P<.001), with an overall random effect size of 0.72 (P<.001, Q=127.3, I² =91.36). One study aimed to improve maternal mental health (eg, reducing postpartum stress and depression), and the effect size was 0.84 (P<.001). Two studies aimed to improve knowledge about birth preparedness and infant feeding, and the effect size was 0.8 (P=.04, Q=3.55, I² =71.82).

We investigated factors that may moderate the program effectiveness, including whether the intervention included interactive treatment content, use of professional consultation or not, the type of technology used, use of a wearable device or not, and whether the participants were overweight or obese. As shown in Table 3, the use of a wearable device to track physical activity in interventions aiming at weight management was a significant moderator (Q_b =5.91, P=.02). Using such a device resulted in a larger pooled effect size (d=0.97). Sample size was also a significant moderator of the effect size (Q_b =7.38, P=.007). Studies with smaller sample sizes resulted in a larger pooled effect size (d=1.13).

Moderator analyses in Table 3 show that interactive content and professional consultation were not significant moderators (Q_b =1.5, P=.22). The effects of interventions providing interactive treatment content and involving professional consultations were not significantly better compared with the interventions without these components. The interventions using social networking (d=0.67) and health and fitness mobile phone apps (d=0.77) were also similarly effective as the result of the Q test was insignificant.

A funnel plot was used to examine publication bias of our meta-analysis, as shown in Figure 4. Almost half of the studies used a larger sample (n>100), and they concentrated around the top of the funnel plot. However, there were more studies on the right side of the mean effect size, especially studies with smaller samples, which made the funnel plot asymmetric. This means that positive results of the interventions based on small sample sizes were more likely to be published. Therefore, there is evidence that publication bias exists.

The use of social media and mHealth apps has been increasing in pregnancy care because of the low cost and their easy access regardless of time and geographic location [40]. This meta-analysis synthesized the findings of 15 RCT studies conducted in different countries and regions and provides evidence on the effectiveness of these technology-based interventions in providing health care services to pregnant and postpartum women. Moderate to large effect sizes were found in regard to different health outcomes including maternal physical health, mental health, and knowledge about pregnancy. In addition, we investigated the characteristics and components of the interventions that may affect program effectiveness.

Pregnancy is a life-changing experience with risks of excessive weight gain and obesity [36]. Postpartum weight retention is a prevalent problem among US women, especially within racial and ethnic minorities [17]. However, it can be difficult for pregnant or postpartum women to manage their weight by increasing physical activity or changing dietary intake because of limited resources in health care or a low level of engagement in health management [16,41]. This meta-analysis finds that there was a moderate effect in maternal weight control and maintaining optimal body composition by promoting lifestyle change and self-monitoring via mHealth apps and social media.

Similarly to pregnancy weight control, GDM control also relies on self-monitoring the change of unhealthy lifestyle and listening to clinical decisions, and patient compliance is particularly important [42,43]. The results of the meta-analysis show that mHealth apps and social media were also effective for pregnant women with GDM, with a small to moderate effect size. As participants in the intervention group only had half of the clinic visits compared with participants in the standard care group in the research of Mackillop et al [42], we estimate that the intervention can be more effective with the same number of clinical visits.

An mHealth app was found very effective in asthma control during pregnancy [35]. However, this study is just an initial step toward understanding the effect of the social media and mHealth apps in physical health outcomes other than weight or GDM control. The result of a case-control study showed that the mHealth app can be effective in urinary incontinence management during pregnancy [44]. We can estimate that mHealth apps can be applied in other health services. To conclude, the positive and significant effects demonstrate that lifestyle intervention using advanced technology can be effective in improving maternal health.

Psychological interventions delivered via mobile phones have been found effective in reducing depression and anxiety in existing meta-analyses [45,46]. However, whether the interventions are effective among pregnant or postpartum women is unclear. This study provided additional evidence that mHealth apps and social media can be useful in reducing pregnancy-related stress and depression. However, there was only one RCT study examining the effectiveness in maternal mental health outcomes [37].

Social media and mobile phone apps are becoming increasingly popular among pregnant women and their partners to access health knowledge and learn to identify risk behaviors and danger signs during pregnancy. The findings from two RCT studies [33,34] demonstrate the usefulness of the intervention programs to prepare the participants to become mothers. A pretest-posttest study found that providing information about maternal and infant care via the mHealth app can reduce maternal stress during pregnancy [47]. Therefore, with the improvement in health knowledge, maternal mental health may also improve.

Finally, we were interested in investigating whether intervention characteristics or components can affect their effectiveness. Interventions using social media and interventions using mobile phone apps resulted in similar effect sizes. In addition, it seems to make little difference whether interactive treatment content or professional consultation was provided in the intervention. One explanation may be that all apps have a reminder function and provide health information, which is similar to some functions of the interactive treatment and professional consultation. Another possible explanation may be that the usefulness of the interventions is more likely to rely on women’s self-monitoring. Therefore, different forms of mHealth apps and social media providing pregnancy care may have similar benefits. Interventions without interactive content or professional consultation can be more cost effective.

The results of moderator analyses also showed that using wearable devices to track participants’ physical activities has the potential to enhance program effectiveness in weight control during the prenatal or postnatal periods. The use of wearable devices may be a good way to improve self-monitoring. Another moderator variable that significantly contributed to the variance in the effect size was sample size. Interventions with smaller sample sizes seem to be more effective, whereas interventions with larger samples were less effective. Olson et al [38] and Dodd et al [28] argued that in their study, the similar contents of intervention and control groups or the low use of self-monitoring tools in the intervention may explain the low program effectiveness. However, the absence of small studies with small effect sizes also indicates publication bias among the studies in this meta-analysis.

Because there was a lack of quantitative integration of the evidence on effectiveness of the social media and mHealth apps, further investigation was recommended before the implementation of the intervention [24]. This study includes rigorous studies that offer high-quality evidence. Our review is the first meta-analysis evaluating program effectiveness through credible statistical analyses.

There are several limitations to this study. First, the presence of publication bias indicates that more studies need to be included, as the studies included in the meta-analysis were more likely to report larger effect sizes. Second, as most studies included provided only limited details about participant profiles, several factors cannot be examined through moderator analysis such as participant socioeconomic status and health status.

Although the use of advanced technology in pregnancy care has been increasing in recent years and there are promising results in improving maternal health outcomes, research in this area is still in its early stages. First, more large-scale clinical trials are suggested in future studies. This is because interventions with smaller sample sizes are more likely to report larger effect sizes, which can result in skewed distribution of effect sizes in the meta-analysis. Also, more studies are suggested to be included in future reviews. Second, because interventions included in this review were used predominantly for managing health problems, the effectiveness in improving mental health of pregnant and postpartum women needs to be examined. Third, cost effectiveness could be an important feature of the use of mHealth apps and social media in pregnancy care [42]; however, it was not examined in most studies. Therefore, cost analysis is necessary in future studies.

This meta-analysis of the effectiveness of social media and mHealth apps has several implications for future practice. First, interventions with the use of social media and mHealth apps can be effective in promoting maternal well-being. The positive effects in developing countries such as Indonesia and China imply that the use of mobile technologies in pregnancy care can be less restricted by social and economic development. Social media and mHealth apps can be widely adopted in different areas and have greater public health impact.

Second, the study of Santoso et al [34] demonstrated that fathers can also be positively involved in pregnancy care and birth preparedness by using social media and health apps. The inclusion of fathers could improve health outcomes for the whole family [1]. Future practice should consider attracting fathers to use the related services.

Third, the use of mHealth apps was poor among participants in some interventions, which may lead to low effectiveness [28,38]. Therefore, it is important for researchers, service providers, and app developers to consider how to increase the use of interventions and customer stickiness. It is also necessary to find useful ways to improve participant self-monitoring.

Fourth, even though this review included a number of mHealth apps, most commonly used mHealth apps are commercial and the credibility of their information is unknown [23]. Therefore, it is necessary to examine the quality and effectiveness of their services. Evidence-based mHealth apps and social media interventions for pregnant women are recommended in the practice.

Social media and mHealth apps are increasingly used in pregnancy care with emerging promising findings. In this meta-analysis, we found the interventions were useful with moderate to large effect sizes in regard to maternal health, mental health, and knowledge about pregnancy. We conclude that social media and mHealth apps have the potential to be widely used in improving maternal well-being during the prenatal and postnatal periods. More large-scale clinical trials with comprehensive aims are suggested for future studies.