Despite progress in improving global maternal mortality, it remains unacceptably high, particularly in low- and middle-income countries (LMICs) [1]. Reducing complications during and following pregnancy and childbirth, when most complications occur, can reduce or prevent maternal mortality. Skilled care during and following pregnancy and childbirth could reduce complications and may result in preventing maternal deaths. Studies conducted in Tanzania and Ethiopia have confirmed the ability of antenatal care (ANC) and skilled birth attendance (SBA) during labor and delivery to reduce maternal mortality [2-4].

Globally, most pregnant women have access to ANC with a skilled health professional (eg, physician, nurse, or midwife) at least once, but only 65% receive the World Health Organization–recommended number of at least four ANC visits, and 81% of births occur with the assistance of skilled health personnel [5]. Although there has been a significant improvement in the coverage for SBA and facility delivery in the last decade, millions of births still occur annually without any assistance from a skilled health professional [6]. Several factors prevent pregnant women from receiving the care provided by skilled health personnel during pregnancy and childbirth, such as lack of information, limited preventive health education, limited access to maternal health services owing to poverty or distance factors, poor administration, shortage of health care professionals, and inadequate or poor-quality services [7,8].

In LMICs, only approximately half of pregnant women receive 4 or more ANC visits, and the rate of skilled delivery care including SBA and facility delivery is relatively poor [5]. Moreover, the lowest levels of ANC and skilled delivery coverage are observed in regions where maternal mortality remains excessively high. For instance, the coverage for 4 or more ANC visits was 49% in South Asia and 52% in sub-Saharan Africa [5]. In terms of SBA at the time of delivery, the coverage was 60% in sub-Saharan Africa and 77% in Southern Asia, whereas other World Health Organization regions have achieved universal coverage [5]. The poor coverage of ANC and SBA accounted for the higher maternal mortality in these regions. Hence, a faster pace of progress is required to improve the coverage of ANC and skilled delivery in these high-burden regions.

In the last decade, mobile phone coverage has rapidly increased worldwide. The International Telecommunication Union reported that global mobile phone subscriptions crossed >7 billion in 2015, and mobile phone penetration reached over 90% in LMICs [9]. Thus, mobile phone penetration has the potential to strengthen existing health care service utilization, particularly ANC, SBA, and facility delivery services, in resource-limited settings in a cost-effective manner. Mobile health (mHealth) interventions are becoming more widespread in LMICs because the technology involved is more rapid and accessible than internet access. Labrique et al [10] reported on 12 mHealth applications that could respond to various health issues. Most mHealth interventions are designed to promote behavior change in patients or health personnel by providing health care reminders, health advice, health education, health information and facilitating referral, or access to health facilities or point-of-care remote consultation. Numerous models of mHealth interventions have been used to support pregnant women during and following pregnancy and childbirth in LMICs [11]. Previous studies have reported that mHealth interventions may be capable of and effective in improving essential maternal health care service utilization in LMICs [12-21]. Most of the systematic reviews narratively synthesized the available literature and reported a great potential for mHealth interventions to change maternal health care–seeking behaviors and showed a positive effect on improving ANC, SBA, postnatal care, or childhood immunization [12,14,15,18,22]. However, most evidence comes from observational studies and pilot or small-scale mHealth intervention studies, and researchers have expressed concerns about the study quality [12,14,15,18,22].

Although mHealth interventions have shown a great potential for behavior change more broadly, there are relatively few rigorous evaluations assessing the effectiveness of mHealth interventions on the uptake of ANC, SBA, and facility delivery utilization among pregnant women [19,21]. A recent systematic review reported a significantly higher mean of ANC attendance in mHealth interventions than standard care, but the result remains inconclusive because of the higher statistical heterogeneity among the studies; they also summarized randomized controlled trials (RCTs) and non-RCTs together into the meta-analysis [21]. Another systematic review considered RCTs to rigorously evaluate the effectiveness of mHealth interventions, and a meta-analysis based on a limited number of studies found a positive effect on improving 4 or more ANC visits and SBA outcomes [19]. Owing to higher statistical heterogeneity and the limited number of studies for each outcome, the evidence remains inconclusive [19]. To the best of our knowledge, no systematic review has reported on the effectiveness of mHealth interventions in improving facility delivery outcomes. Therefore, a rigorous evaluation of high-level studies is required to assess the effects of mHealth interventions on ANC and skilled delivery care utilization (SBA and facility delivery) in LMICs. This review further summarizes the findings of high-level studies (such as RCTs) and provides a clear direction to health practitioners, researchers, and policy makers.

The objective of this review is to explore and synthesize the effects of mHealth interventions on improving the uptake of ANC visits, SBA at the time of delivery, and facility delivery among healthy pregnant women in resource-limited settings.

The review protocol was registered in the PROSPERO database (CRD42020210813) [23]. The guideline of PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) was followed for reporting this systematic review and meta-analysis [24].

Using a highly sensitive search strategy, we conducted a comprehensive search from inception to October 2020 to identify RCTs, including cluster RCTs, in the following electronic databases: APA PsycINFO, British Nursing Index, CINAHL Plus, Embase, MEDLINE, POPLINE, PubMed, The Cochrane Library, and Web of Science. We used controlled vocabulary and text words for each database. The search terms were grouped into three major categories of interest: participants (pregnant women), interventions (mHealth interventions), and study designs (RCTs). We did not limit our search to language, date, or publication type to include all published studies. Moreover, we checked the reference lists of all the included studies and relevant systematic reviews to identify additional potential studies for inclusion. The details of the search strategies for each database are provided in Multimedia Appendix 1.

Two reviewers (MOR and NY) independently screened the titles and abstracts of all retrieved studies and identified potentially relevant studies using the predefined study eligibility criteria. To assess their eligibility in detail, they independently critiqued all potentially relevant studies during the full-text screening stage. In both stages, disagreements were resolved through discussion or by a third reviewer (YN or EO), when required. We recorded the reasons for exclusion of all studies in the full-text screening stage and reported them in a PRISMA study flow diagram. We used EndNote (Niles Software) reference management software and the Rayyan Qatar Computing Research Institute tool in the study selection process [26].

Two reviewers (MOR and NY) independently assessed the risk of bias of the included studies using the Cochrane risk of bias tool. The tool consists of the following domains: random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, selective outcome reporting, and other bias. For the domain of blinding of participants and personnel, we considered blinding of personnel only as blinding, because the nature of mHealth interventions, may not be possible to study participants. We classified the studies with high, low, and unclear risks of bias based on the Cochrane Handbook [27]. Any discrepancies were solved through discussion or by a third reviewer (YN or EO).

We evaluated certainty of evidence for ANC, SBA, and facility delivery outcome using the Grading of Recommendation, Assessment, Development and Evaluation (GRADE) system [28]. The GRADE system considered the judgment on the following factors while assessing the confidence in the evidence based on RCT studies: study limitations (risk of bias), inconsistency (statistical heterogeneity), indirectness (PICO [participants, interventions, comparisons, and outcomes] and applicability), imprecision (number of events and CIs), and publication bias. On the basis of the judgment of each factor, we classified our evidence as follows: (1) high-certainty evidence (further research is very unlikely to change the confidence of the pooled results), (2) moderate-certainty evidence (further research is likely to have an important impact on the confidence of the pooled results and may change the estimate), (3) low-certainty evidence (further research is extremely likely to have an important impact on the confidence of the pooled results and likely to change the estimate), and (4) very low-certainty evidence (the pooled results have extreme uncertainty) [28]. We used the GRADEpro web-based platform to make a summary of findings table, considering the certainty of evidence.

Two independent reviewers (MOR and NY) extracted a standard set of data, including study characteristics, participant characteristics, description of interventions, and outcome results from each of the selected studies and were cross-checked. As in the study selection process, disagreements were resolved through discussion or by a third reviewer (YN or EO). We reported key characteristics of the included studies in a separate table. The data characteristics included, but were not limited to, author information, year of publication, study location, study setting, study design, study name, number of participants, study year, age of participants, gestational age at recruitment, comparator, types, function, mode and duration of interventions, intervention provider, and reported outcomes with their results.

We narratively synthesized study characteristics, participant characteristics, intervention characteristics and key findings among all included studies. To summarize the effect size of mHealth interventions, we used pairwise inverse-variance random-effects meta-analysis separately for each outcome. While pooling the effect size, we used risk ratios (RRs) because our outcome was dichotomous in nature. If the study provided odds ratios (ORs) and the risk of events in the control group (assumed control risk [ACR]), we converted ORs into RRs by using the formula (OR/[1 – ACR × (1 − OR)]) described in the Cochrane Handbook for Systematic Reviews of Intervention [29]. The results of the meta-analysis are presented in forest plots. In the meta-analysis, we used the estimated effective samples for cluster RCT studies by adjusting their design effect if the studies reported unadjusted data. Heterogeneity was assessed by visual inspection of forest plots or tested using the I² statistic, and we considered an I² value >50% to indicate substantial heterogeneity [29]. If substantial heterogeneity was found in the meta-analyses, we conducted a subgroup analysis based on the direction of interventions (1-way vs 2-way communication) and high baseline coverage of outcomes (80% or more vs <80%) and reported the subgroup-wise pooled estimates for each outcome separately. We used funnel plots and the Egger test to assess publication bias if a meta-analysis includes 10 or more studies. Statistical significance was defined as a P value <.05 for all analyses.

This study did not require ethical approval or consent to participate, as it used data from published studies.

A total of 3085 potentially relevant articles were retrieved from all targeted electronic databases and other resources. After removing duplicates in EndNote, 2335 unique articles underwent initial title and abstract screening. As a result, 67 articles were retained for a detailed assessment of study eligibility. After full-text screening, 57 articles that failed to meet the study eligibility criteria were excluded. The reasons for exclusion are reported in the PRISMA study flow diagram (Figure 1). Finally, 10 articles (9 studies) from all resources were found to be suitable for narrative synthesis. We identified 1 gray article by checking the reference lists of all included studies and relevant systematic reviews, but the full text was not available [30]. As we could not check its quantitative information, we narratively synthesized its results and excluded it from our meta-analysis.

Of all 9 included studies, 6 (67%) assessed the effect of mHealth interventions on improving the uptake of ANC visits [17,30-34], 5 (56%) on SBA during delivery [16,31-33,35], and 3 (33%) on facility delivery outcome [31,36,37]. We performed a meta-analysis for ANC visits and SBA outcomes based on the direction of interventions (1-way or 2-way communication) and for facility delivery based on high coverage of outcomes at baseline (80% or more; Figure 1).

Of all 9 included studies, 6 (67%) were individual RCTs, and 3 (33%) were cluster RCTs (Table 1). Studies were conducted in Brazil [34], China [30], Ethiopia [31], India [32,35], Kenya [33,37], Nigeria [36], and Tanzania [16,17] and were published between 2012 and 2018. A total of 10,348 pregnant women (n=6254, 60.44% in the intervention groups and n=4094, 39.56% in the control groups) participated in the studies. The number of study participants ranged from 116 to 2160 in the intervention groups and from 100 to 1239 in the control groups.

Interventions were carried out through SMS text messaging (5/9, 56% studies), voice calling (1/9, 11% studies), SMS text messaging and mobile voucher (1/9, 11% studies), voice calling and SMS text messaging (1/9, 11% studies), and voice messaging and animation film clips (1/9, 11% studies). Approximately half of the studies used interventions for 1-way communication (4/9, 44% studies) or 2-way communications (5/9, 56% studies). The functions of interventions were categorized into appointment reminder (1/9, 11% studies), health education or advice and appointment reminder (5/9, 56% studies), and health education or preventive health information (3/9, 33% studies). The detailed characteristics of the interventions and the results of the included studies are presented in Table 2.

A summary of the overall risk of bias assessment is presented in Figure 2 and Figure 3. RCT studies generally performed well in their risk of bias for random sequence generation (75% low risk, with the remainder unclear), allocation concealment (100% low risk), blinding outcome assessment (37.5% low risk, with the remainder unclear), incomplete outcome data (87.5% low risk), and selective reporting (100% low risk). A total of 3 studies [32,34,37] had a high risk of bias regarding blinding participants and personnel, and another study [34] also had a high risk of bias for incomplete outcome data. In addition, 1 study [33] had an unclear risk for many items. Overall, most of the studies had a low risk of bias.

We performed an inverse-variance random-effects meta-analysis to summarize the effects of 1-way and 2-way mHealth interventions versus standard care on the uptake of 4 or more ANC visits among pregnant women (Figure 4 and Table 3). A total of 2 studies [31,32] consisting of 1945 pregnant women (n=1206, 62.01% in the intervention group and n=739, 37.99% in the control group) implemented a 1-way mHealth intervention and were included in the meta-analysis. The pooled estimates showed a significantly large risk with a 114% increase in 4 or more ANC visits among pregnant women given a 1-way mHealth intervention, compared with the control group (RR 2.14, 95% CI 1.76-2.60, I²=36%, moderate certainty of evidence).

In total, 3 studies [17,33,34] consisting of 1762 pregnant women (n=664, 37.68% in the intervention group and n=1098, 62.32% in the control group) reported 2-way mHealth intervention, and the pooled estimates from the meta-analysis showed that the risk of 4 or more ANC visits was 17% higher in the 2-way mHealth intervention group than in the control group (RR 1.17, 95% CI 1.08-1.27, I²=59%, low certainty of evidence). Although all studies reported a positive effect of mHealth interventions, regardless of the direction of interventions, on improving the uptake of ANC visit utilization [17,31-34], we observed a significant difference between these subgroups (I²=96.9%), which limited their combination.

An inverse-variance random-effects meta-analysis was performed to pool the effects of 1-way and 2-way mHealth interventions versus standard care on improving the rate of SBA at the time of delivery (Figure 5 and Table 3). In total, 3 studies [31,32,35] comprising 3460 pregnant women (n=2216, 64.05% in the intervention group and n=1244, 35.85% in the control group) reported a 1-way mHealth intervention in which only 1 study [31] found a positive effect of the intervention on SBA during delivery. The effects of 1-way mHealth interventions on SBA during delivery were not clear; however, the effects were pooled in the meta-analysis (RR 1.04, 95% CI 0.97-1.10, I²=73%, very low certainty of evidence).

In total, 2 studies [16,33] involving 1212 pregnant women (n=612, 50.50% in the intervention group and n=600, 49.50% in the control group) implemented 2-way mHealth interventions, and their pooled estimates showed that the proportion of SBA during delivery was 23% higher in the 2-way mHealth intervention group than in the control group (RR 1.23, 95% CI 1.14-1.33, I²=0%, moderate certainty of evidence). Owing to significant subgroup differences in the direction of mHealth interventions (I²=91.3%), we could not combine their effects.

Overall, 3 studies assessed the effects of mHealth interventions on facility delivery outcome [31,36,37]. In addition, 1 study [31] found that the probability of facility delivery was 57% higher in the 1-way mHealth intervention group than in the control group (RR 1.57, 95% CI 1.41-1.75). Another study [36] identified a positive result supporting the effectiveness of 2-way mHealth interventions for improving facility delivery (RR 2.18, 95% CI 1.32-3.60). However, 1 study reported an unclear effect of the mHealth interventions, irrespective of direction, on the uptake of facility delivery [37]. We conducted a subgroup analysis considering high coverage of facility delivery at baseline (80% or more) and found a positive effect of mHealth interventions on improving the uptake of facility delivery in LMICs, where fewer pregnant women used facility delivery at baseline (RR 1.68, 95% CI 1.30-2.19, I²=36%, moderate certainty of evidence); however, the effects were unclear where most pregnant women already used facility delivery (RR 1.01, 95% CI 0.97-1.04, I²=0%, low certainty of evidence). Significant subgroup differences (I²=93.2%) limited the combination of these effects (Figure 6 and Table 3).

Significant subgroup differences (P<.001) were found in all meta-analyses, which limited the combination of the effects of all included studies (Figures 4-6). It included an estimated 96.9% subgroup variance for 4 or more ANC visits, 91.3% for SBA at the time of delivery, and 93.2% for facility delivery.

Although we planned to assess the publication bias of our meta-analyses, we did not perform the analysis because of the limited number of studies.

This meta-analysis identified a statistically significant positive effect of mHealth interventions, regardless of the direction of interventions (1-way or 2-way communications), on improving ANC care utilization of healthy pregnant women in LMICs. Only 2-way mHealth interventions were effective in improving SBA during delivery, but the effects were unclear for 1-way mHealth interventions compared with standard care. For facility delivery among healthy pregnant women, the interventions were effective in settings where fewer pregnant women used facility delivery. Most studies that combined multiple mHealth interventions were implemented in Brazil, China, Ethiopia, India, Kenya, Nigeria, and Tanzania. The functions or directions of interventions varied among the included studies, such as 1-way or 2-way communication, appointment reminder, or health advice. All studies, except one [34], had low concerns of their methodological qualities. However, high statistical heterogeneity limited the combination of subgroups in all the meta-analyses.

These findings are consistent with a systematic review reporting that mHealth interventions had a positive effect and resulted in a 43% increase in the uptake of recommended ANC visits among pregnant women [13]. In another systematic review conducted by Wagnew et al [19], SMS text messaging had positive effects on the uptake of 4 or more ANC visits (OR 2.74, 95% CI 1.41-5.32) and SBA (OR 1.82, 95% CI 1.33-2.49) in LMICs [19], which strongly supports our findings. Although the availability of high-level evidence such as RCTs on mHealth interventions targeting healthy pregnant women’s health care utilization is still very limited, our findings have generated promising results regarding the positive effects of mHealth interventions, regardless of their directions, on improving recommended ANC utilization, SBA during delivery, and facility delivery in resource-limited countries. The findings are also consistent with other systematic reviews that found that mHealth tools are effective in influencing maternal and child health service utilization by enhancing the uptake of recommended ANC and postnatal care services, including SBA, at the time of delivery and institutional delivery [12,14,18].

The effects of 1-way mHealth interventions on SBA at the time of delivery were not clear in the 2 included studies [32,35], which was reflected in our pooled results. Our meta-analysis identified a significant difference in the likelihood of SBA during delivery, which was higher in the 2-way mHealth intervention group than in the control group. Our findings support the effectiveness of mHealth interventions in improving facility delivery in settings where fewer pregnant women use the service. Consistent with our findings, a systematic review conducted by Colaci et al [38] reported that mHealth interventions offered an opportunity to increase the acceptability of prenatal and obstetric care, including SBA at the time of delivery. This is because mHealth interventions, either health care reminders or health advice, boost self-efficacy and access to care among pregnant women and create closer interactions with health service providers. Furthermore, mHealth interventions have been used as appointment reminders and can provide basic health information, notably throughout the pregnancy period. As novel and more cost-effective systems are being sought to promote health care utilization in underserved areas, particularly in remote and rural areas, this intervention offers a potential cost-effective solution. This study strongly supports the use of mHealth interventions, either 1-way or 2-way communication, to enhance the uptake of maternal health care services such as ANC, SBA, or facility delivery, by changing health care behavior among pregnant women in similar settings.

This review had several strengths. To the best of our knowledge, this is the first comprehensive review that conducted meta-analyses for different subgroups, even in similar settings, for each outcome, such as the direction of interventions (1-way or 2-way communication) and high coverage of outcomes at baseline, and identified some novel findings not seen in previous studies. For example, 2-way (not 1-way) mHealth interventions likely result in an increase in SBA during delivery in resource-limited settings. Likewise, mHealth interventions likely result in an increase in facility delivery in LMICs where fewer pregnant women use the service but may not increase in settings where most pregnant women already use facility delivery. Interestingly, the certainty of the evidence was moderate, indicating that our estimates were likely to be close to the true effect.

Second, we conducted a comprehensive search of electronic databases without any limit on language, date, or type of publication. We also checked the reference lists of the included studies and other relevant systematic reviews to avoid missing any potentially relevant studies, and identified 1 additional study. Third, this study considered high-level studies such as RCTs and cluster RCTs and performed meta-analyses well. In the meta-analyses, we used the estimated effective samples for cluster RCT studies; however, the studies did not adjust their design effect. In addition, with the larger sample size, we enhanced the statistical power to provide more precise and reliable effect estimates.

Despite the positive effects of mHealth interventions reported in our meta-analyses, this review had several limitations. One of the important limitations is the limited number of RCT studies included in the meta-analyses (8 studies). All included studies were reported from only 7 LMICs, and some used small sample sizes that may compromise representativeness. In addition, approximately half of the studies tried to combine multiple mHealth interventions, making it difficult to understand the extent to which each intervention contributed to the observed results. For instance, a study assessed the effectiveness of a combined intervention of mobile SMS text messaging and mobile voucher [16,17], another studied a voice messaging and animation film clip intervention [35] and a mobile phone call and SMS text messaging intervention [32]. Likewise, some studies tried to combine multiple functions of mHealth interventions (eg, health education or advice and appointment reminder), which made it difficult to determine the extent to which each function of intervention contributed to the reported outcomes [16,17,32,33,36].

Although the mHealth interventions were effective in improving facility delivery in the settings where fewer pregnant women used the service, we need to interpret these results with caution because it is not clear if the benefits presented regarding facility delivery are a function of the mHealth interventions or greater health literacy and knowledge of navigating the local health care system among users and because it is also not clear if the increase in ANC and SBA are due to the mHealth technology or access to health care. With the mHealth interventions, we also need to consider other factors such as sociocultural norms and beliefs during perinatal periods, perinatal care availability and resources, or other structural factors to improve ANC, SBA, and facility delivery utilization among pregnant women.

We performed an inverse-variance random-effects meta-analysis to summarize the effect of mHealth interventions, irrespective of their directions or high coverage of outcomes at baseline, on the uptake of ANC visits, SBA, and facility delivery utilization among pregnant women in LMICs; however, we observed a significant difference between these subgroups that limited their combination. Within the subgroups, we did not observe considerable heterogeneity among the included studies.

mHealth interventions (1-way or 2-way communication) may contribute toward improving maternal health care–seeking behavior throughout the pregnancy cycle. Public health researchers, practitioners, and policy makers should consider such interventions in resource-limited settings. It is reasonable to use mHealth interventions in resource-limited settings, as this study found a positive effect of different forms of intervention on the uptake of ANC visits, SBA, and facility delivery utilization. This study could play an important role in addressing the Sustainable Development Goal of Maternal and Child Health, as it provides insights and evidence-informed recommendations for the utilization of different forms of mHealth interventions in addressing maternal health care challenges in LMICs. However, owing to the limited number of RCT studies that met the study eligibility criteria in this review, further large-scale RCTs are suggested to be implemented in resource-limited countries, particularly where service utilization is quite low among pregnant women. In recent times, most LMICs have recognized the need for appropriate technology use strategies to promote the utilization of their existing health systems, which also justifies the necessity of further evidence of technology-based health care interventions such as mHealth interventions.

Although mHealth interventions (1-way or 2-way communication) can improve the uptake of ANC and skilled delivery care utilization, more rigorous evaluations are required to strengthen the evidence of different forms of mHealth interventions for improving the existing health care utilization among healthy pregnant women in LMICs. This systematic review and meta-analysis will help public health researchers or policy makers in designing and implementing mHealth interventions in resource-limited settings, as this study identified some novel findings not found in previous reviews.