Assessing the Effect of mHealth Interventions in Improving Maternal and Neonatal Care in Low- and Middle-Income Countries: A Systematic Review

Introduction Maternal and neonatal mortality remains high in many low- and middle-income countries (LMIC). Availability and use of mobile phones is increasing rapidly with 90% of persons in developing countries having a mobile-cellular subscription. Mobile health (mHealth) interventions have been proposed as effective solutions to improve maternal and neonatal health. This systematic review assessed the effect of mHealth interventions that support pregnant women during the antenatal, birth and postnatal period in LMIC. Methods The review was registered with Prospero (CRD42014010292). Six databases were searched from June 2014–April 2015, accompanied by grey literature search using pre-defined search terms linked to pregnant women in LMIC and mHealth. Quality of articles was assessed with an adapted Cochrane Risk of Bias Tool. Because of heterogeneity in outcomes, settings and study designs a narrative synthesis of quantitative results of intervention studies on maternal outcomes, neonatal outcomes, service utilization, and healthy pregnancy education was conducted. Qualitative and quantitative results were synthesized with a strengths, weaknesses, opportunities, and threats analysis. Results In total, 3777 articles were found, of which 27 studies were included: twelve intervention studies and fifteen descriptive studies. mHealth interventions targeted at pregnant women increased maternal and neonatal service utilization shown through increased antenatal care attendance, facility-service utilization, skilled attendance at birth, and vaccination rates. Few articles assessed the effect on maternal or neonatal health outcomes, with inconsistent results. Conclusion mHealth interventions may be effective solutions to improve maternal and neonatal service utilization. Further studies assessing mHealth’s impact on maternal and neonatal outcomes are recommended. The emerging trend of strong experimental research designs with randomized controlled trials, combined with feasibility research, government involvement and integration of mHealth interventions into the healthcare system is encouraging and can pave the way to improved decision making on best practice implementation of mHealth interventions.

increased maternal and neonatal service utilization shown through increased antenatal care attendance, facility-service utilization, skilled attendance at birth, and vaccination rates. Few articles assessed the effect on maternal or neonatal health outcomes, with inconsistent results.
Conclusion mHealth interventions may be effective solutions to improve maternal and neonatal service utilization. Further studies assessing mHealth's impact on maternal and neonatal outcomes are recommended. The emerging trend of strong experimental research designs with randomized controlled trials, combined with feasibility research, government involvement and integration of mHealth interventions into the healthcare system is encouraging and can pave the way to improved decision making on best practice implementation of mHealth interventions.

Background
The availability and use of mobile phones is increasing rapidly in low-and middle-income countries (LMIC) [1][2][3]. In 2014, 90% of persons in developing countries have a mobile-cellular subscription (pre-paid and post-paid), 89% in the Asia-Pacific region and 69% in Africa [2]. These countries are responsible for more than 75% of mobile-cellular subscriptions globally [2]. The wide availability of mobile phones and their ease of use have given rise to the field of mobile health (mHealth), in which mobile phones and tablets support medical and public health practice [3][4][5][6]. mHealth interventions can be used to provide varying functions: educational information, support, reminders, emergency response, and monitoring [7]. In LMIC this means mHealth could reduce time, distance, and cost of information delivery, and thus overcome issues of inadequate financing, poor access to information, and limited human resources [8]. mHealth interventions are being used for health care strengthening by governments, nongovernmental organizations (NGOs), donors, multilateral agencies and corporations in LMIC [3,6].
One of the key-areas addressed by mHealth interventions is the support of pregnant women during the antenatal, birth and postnatal period, in order to tackle high maternal and neonatal mortality [3]. Maternal and neonatal mortality remain high in LMIC despite progress in Millennium Development Goals (MDGs) 4 and 5 [9]. Of the 289,000 maternal deaths in 2013, 286,000 occurred in developing regions [10]. Sub-Saharan Africa (SSA) accounted for 62% of all maternal deaths in 2014 [10]. Similarly, LMIC account for the majority of the 2,612,100 neonatal deaths worldwide [11,12],which is approximately 40% of the deaths of children under five [1].
Between 2011 and 2013, Noordam et al., Tamrat and Kachnowski, and Philbrick published reviews assessing the effectiveness of mHealth interventions targeting maternal and neonatal care [13][14][15]. Given the relatively emerging field of research and the wide interest in mHealth interventions to improve maternal and neonatal health, a substantial number of studies were published since. In addition, the reviews had quality limitations.
The increased drive to develop and scale-up mHealth interventions, demands availability of robust evidence of the effect [5].Therefore, the main objective of this study was to conduct a systematic review to assess the effect of mHealth interventions targeted at pregnant women to improve maternal and neonatal care in LMIC.

Protocol and registration
This review is part of a larger systematic review which also included mHealth interventions focussed on midwives and health care providers bestowing maternal and neonatal care. It was registered with the PROSPERO review of registry for systematic reviews (CRD42014010292), and is based on the guidelines provided by PRISMA [16](S1 File).

Eligibility criteria
Studies focussing on the domain of pregnant women during antenatal, labour and postnatal care up to 28 days postpartum in LMIC, and the determinant mHealth were eligible for inclusion. LMIC were defined according to the World Bank Classification [17]. mHealth was defined as a medical and public health practice supported by mobile phones and tablets, making use of text, audio, images, video or coded data in the form of short messaging services (SMS), voice SMS, applications accessible via general packet radio service (GPRS), global positioning system (GPS), third and fourth generation mobile telecommunications, and Bluetooth. mHealth supports the exchange of health related information and provides varying functions: educational information, support, reminders, emergency response, and monitoring [7]. Outcomes were not pre-specified in the search or eligibility criteria given the interest in any outcomes related to our domain and intervention. Studies were excluded when their outcomes did not address outcomes within the antenatal, labour and postnatal period up to 28 days postpartum.
Articles were eligible for inclusion when written in English, Dutch, French, German or Spanish, contained the pre-defined domain and determinant, and were a primary study. Data of results published multiple times were used only once, based on the most comprehensive publication. When an intervention was discussed in more than one article, the study was counted as one, but the outcomes of the different articles shown separately. Articles that included mHealth in a package alongside other non-mHealth interventions, were considered separately. Exclusion criteria included articles not matching the domain and determinant, reports, proceedings, conference abstracts, project protocols and secondary analyses. Interventions relating to the termination of pregnancy were excluded due to the focus on maternity services. Family planning was excluded when it did not start within the post-partum period. Interventions using radio were also excluded as they fell outside the scope of the definition of mHealth. Personal digital assistants were only included when their use in the intervention fit the given definition of mHealth. also included and approached. Reference lists of included studies were screened for additional articles that fit the eligibility criteria.

Study selection
All duplicate articles were removed manually using Endnote (version 11). Screening based on title and abstract was done independently by two reviewers for the database search (SFVS and AB) and by four reviewers for the grey literature search (ASM, MV, SFVS and AB). Any discrepancies between the two reviewers in this process were discussed with the other review team members until consensus was reached and full text was accessed if necessary for further clarification. When access to full-text articles was not available, authors were contacted once. If no reply was received within a month, the study was excluded.

Data collection process and data items
Data extraction of the database articles was done by one reviewer (SFVS) and four reviewers extracted data of the grey literature articles (ASM, JB, MV and SFVS). None were blinded for journal or author details according to a standardized data extraction form. Other authors double-checked the data extraction for accuracy (AB, KKG and MC). Information was extracted on: study design [19,20], research methods, location/healthcare setting, target population/size, mHealth function (educational, monitoring, reminder, communication and support, and emergency medical response system), form of mHealth (unidirectional text/voice messaging, direct two-way communication, multidirectional text messaging, unidirectional telephone counselling), results per classification (maternal outcomes, neonatal outcomes, maternal and neonatal service utilization, and education), strengths, weaknesses, opportunities and threats of the intervention. Classification of mHealth functions and forms was determined after identification of emerging themes from the results [21]. Where possible, the summary measures risk ratio and odds ratio were used for results. Lack of clarity during the extraction process was resolved by consulting members of the research team (AB, ASM, JB, KKG, MC and MV) until consensus was reached. In case of incomplete data, one attempt was made to contact the corresponding author or organization by e-mail.

Risk of bias assessment
The intervention articles were assessed for quality according to an adaptation of the Cochrane Risk of Bias Tool for intervention studies [19]. Bias was assessed on the sequence generation, allocation concealment, selection process of the study population, completeness of data (e.g. number of drop-outs), origin of the data (measurements performed by authors or database research), blinding of the researchers or clinicians, the presence of a clear definition of the outcomes that were used, and whether confounders were taken into account. Bias risk was assigned as either one of three levels (low/high/or unclear risk) and taken into consideration in determining the strength of the conclusion as described in the discussion section. The quality assessment criteria are available in the S1 Table. Risk of bias could not be assessed across studies through a funnel plot or Egger's tests.
The quantitative results of intervention studies were summarized in an evidence table according to study type: randomized controlled trial (RCT) and non-randomized study (NRCT) [20]. A narrative synthesis of the results per classification was given . The descriptive studies were  summarized in a similar evidence table to the intervention studies. A narrative synthesis of qualitative information was performed with an analysis of the strengths, weaknesses, opportunities, and threats (SWOT) of all the included studies for the domains accessibility, acceptability, and usability. The SWOT analysis was done by one of the reviewers (SFVS, supported by JLB, MV and ASM). The rationale behind conducting a SWOT analysis was the analytical framework that it provides for the identification of internal (strengths and weaknesses) and external factors (opportunities and threats) that influence the effect and possible scale-up of mHealth interventions [18,22].

Results
The search yielded a total of 3214 articles (3659 through database searching and 117 through the grey literature search) after removal of duplicates and 1 article through snowballing (Fig 1). After full-text screening, a total of 29 articles were included. Of these, fourteen were intervention studies and fifteen were descriptive studies. Of the intervention studies, three belonged to one intervention [23][24][25] and were counted as one intervention study, leaving a total of twelve intervention studies included. Table 1 Tables 2  and 3 give an overview of the characteristics and results of the intervention and descriptive studies respectively. All studies were included for the SWOT analysis (Table 4).

Overall risk of bias assessment of intervention studies
A summary of the overall risk assessment is shown in Fig 3 and the quality assessment of the intervention studies are shown in Table 5 with the detailed version available in S2 Table. Intervention articles generally performed well in their risk of bias for the selection of study population (66% low risk), completeness of data (83% low risk), clear definition of outcome (100% low risk) and confounders (50% low risk, with the remainder unclear). A number of studies displayed high risk of bias in sequence generation (58%), allocation concealment (41%), or origin of data (25%).
Of the RCTs, Lund et al. and Tahir and Al-Sadat had a low risk for all items [23][24][25]. Khorshid et al. and Jareethum et al. had an unclear risk regarding allocation concealment and the blinding of the researcher. The latter also had an unclear risk concerning confounders [27,28]. Lau et al. and Ross et al. had high risks of bias in their randomization and selection of study population [29,44]. The studies with a non-randomized study design generally lost quality with regards to sequence generation and allocation concealment and the selection of the study population. Oyeyemi and Wynn, Watkins et al. and Kaewkungwal et al. displayed a high risk in the origin of data [33,37,47]. The study of Pathak had an unclear risk for many items [32]. An unclear risk was often found for the item of confounders [26,32,37,38,47].

Narrative synthesis of quantitative results
Maternal and neonatal service utilization. All studies addressing maternal and neonatal service utilization showed significant increases.
For maternal service utilization, several studies showed positive effects on antenatal care (ANC) attendance. The Wired-Mothers intervention of Lund et al. more than doubled the odds of a woman receiving four or more ANC visits (OR 2.39, 95% CI 1.03 to 5.55) [24]. The pre-post intervention study in Thailand of Kaewkungwal et al. also showed higher ANC attendance rates after reminders were sent via text messaging (ANC visits: OR 2.97, 95% CI 1.60 to 5.54) [33]. The Chipatala Cha Pa Foni program in Malawi by Watkins et al. combined a toll free case management hotline and unidirectional text and voice messaging to provide protocol-based health information and advice on appropriate care seeking, health practices, referrals and reminders. They found both increased ANC and postnatal care attendance (PNC) [47]. In Sierra Leone, direct two-way communication was set up amongst healthcare workers, between healthcare workers and pregnant women, and between healthcare workers and traditional birth attendants, to improve maternal and newborn health service utilization. The intervention showed a positive net effect on facility-based service utilization for the following indicators: first and fourth ANC visit (0.7 and 11.3%-points, facility delivery (8.2%-points), and first, second and third PNC visit (10.1, 10.6 and 14.9%-points) [38]. The effect decreased or became negative, however, when the chiefdom containing the district headquarter town (which is relatively urban and has relatively better and a great variety of services available) was controlled for [38]. Oyeyemi and Wynn found a significantly higher facility utilization rate within the area in Nigeria taking part in a mHealth intervention (43.4% versus 36.7%, p = 0.0001) [37]. They defined facility utilization rate as the number of deliveries in a particular health facility to the number of ANC registrations in that same facility [37]. Skilled attendance at birth was increased in the study by Lund et al. (60% in the intervention group compared to 47% in the control group) [25], especially for women in an urban area (OR 5.73, 95% CI 1.51 to 22.81) [25].
Two studies in Thailand addressed the effect of mHealth interventions on the emotional aspects of pregnancy. Jareethum et al. observed significantly higher satisfaction scores in the antenatal and perinatal period and high confidence scores and low anxiety levels when educational text messages were sent twice per week [27]. The study of Ross et al. showed a significant decrease in depressive symptoms amongst the HIV-positive pregnant women in the intervention group which received educational and emotional support via telephone [44].
Regarding neonatal service utilization, the pre-post intervention study of Kaewkungwal et al. showed that reminders via text messaging resulted in a higher services-on-time rates of the extended programme on immunization (EPI) (OR 1.48, 95% CI 1.09 to 2.03) [33]. Similar results were observed by Pathak, in a vaccination project in India with unidirectional text messaging to remind mothers to take their newborn for vaccination. The pilot showed a high success rate: 95% rate of the first dose of BCG/HBV/OPV (second dose rate of 98% and third dose rate of 100%) compared to 60% in total at baseline [32].
Maternal outcomes. No studies reported on maternal mortality or severe acute maternal morbidity. Only Oyeyemi and Wynn reported on cases of maternal deaths. They compared the effect of distributing Closed-Users'Group phones to pregnant women and health workers through which they could communicate with each other and among themselves for free in one area to another area in Nigeria and found no significant difference between the two regarding cases of maternal deaths [37].
Neonatal outcomes. Lund et al. observed a significant effect on perinatal mortality in their study conducted in Zanzibar. Their Wired Mothers intervention combined unidirectional text messaging and direct two-way communication in a free call voucher system to provide education on pregnancy, reminders for antenatal care visits and an emergency medical response system. They found a significant decrease in the perinatal mortality rate of 50% (OR 0.50, 95% CI 0.27 to 0.90) [25]. The total perinatal mortality rate based on stillbirth and neonatal mortality was 27 per 1000 births, 19 per 1000 births in the intervention group compared to 36 per 1000 births in the control group [25]. Jareethum et al. who assessed the effect of two educational text messages sent weekly in Thailand, found no differences for infant birth weight and preterm delivery [27].
Education about a healthy pregnancy. The studies assessing antenatal health knowledge showed varying results. Lau et al. found no statistical difference in antenatal health knowledge assessed by nine questions between the pregnant women receiving educational text messages and those who did not receive messages [29]. The SMS were, however, mentioned as the main source of antenatal health knowledge by participants (98%) [29]. Datta et al. did find a significant increase in respondent's knowledge after the intervention on several maternal and  [23][24][25]27,44] Uncertainty whether information of message is received correctly [26] (D: [31]) (G: [36]) Combination of mHealth forms allowing different needs to be addressed [38,46] Privacy is not always guaranteed [44] (D: [31]) Simple mobile phone technology that is easy to use [23][24][25]33] (D: [30]) Uncertainty whether message is received (G: [36]) Question-answer system can provide more interaction when SMS messaging is used (G: [39]) Dependent on donor funding, sustainability (G: [39]) Developing the intervention locally facilitates implementation as it builds on pre-existing knowledge of mobile phone use [23][24][25] Voice messages can be missed and are difficult to store for future reference (in comparison to SMS) (G: [34]) Target audience represents a significant group (i.e. all pregnant women) which offers opportunities for advertising/ revenue to be generated to make it sustainable (G: [42]) Text limits prevent lengthy messages being sent [26] Text limits require skills to design useful health messages fitting the limit (G: [34]) Durability of phones is not always sufficient (G: [34]) neonatal topics in their pre-post intervention study [26]. Watkins et al. in Malawi showed that their participant's baseline knowledge indicator was high at baseline and did not differ significantly between women residing in the intervention and control catchment area, except for women living at a greater distance from the health centre [47]. They observed a positive effect in home-based practices: including use of bed nets during pregnancy and for their children (25% increase for both) and the number of children breastfed within one hour after birth (15% increase). Tahir and Al-Sadat and Jiang et al. both looked into the effect of unidirectional telephone counselling providing education on breast-and infant feeding practices, in order to encourage exclusive breastfeeding. Tahir and Al-Sadat found that 84.3% of the women in the intervention group breastfed exclusively one month postpartum, compared to 74.7% in the control group with a significant odds ratio of 1.83. However, when adjusted for significant factors relating to exclusive breastfeeding, the odds ratio was no longer significant (OR 1.63, 95% CI 0.82 to 3.22) [46]. Jiang et al. found median durations of exclusive breastfeeding and exclusive breastfeeding rates to be higher in the intervention group (median durations: 11.4 weeks compared to 8.9 weeks (p<0.001); exclusive breastfeeding rates: 15.1% in the intervention compared to 6.3% (adjusted OR 2.67; 95% CI 1.45 to 4.91)) in the control group. They also observed a decreased
Strengths of mHealth interventions on usability included the ease of use [23][24][25]30,33], flexibility in use and adaptability to the time in pregnancy [23][24][25]27,44], and development within the local context [23][24][25]. Weaknesses of mHealth interventions included the uncertainty whether a text is received [26,31,36] and the limited length of messages [34]. A combination of the different forms of mHealth, such as educational text messaging and direct two-way communication for emergencies, provide an opportunity for future interventions [38,46].

Discussion
This systematic reviewed suggests that mHealth interventions targeted at pregnant women can increase antenatal and postnatal care attendance, facility-based deliveries, skilled attendance at birth, and vaccination rates. No consistent effects of mHealth interventions on maternal and neonatal health outcomes were observed, though single studies describe benefits regarding reduced perinatal mortality and improved breastfeeding practices. Other studied health outcomes such as anaemia, gestational age at delivery, mode of delivery, neonatal birth weight, preterm delivery, stillbirth and neonatal mortality were not significantly affected by mHealth interventions. Inconclusive findings were observed on the effect of mHealth on health knowledge of pregnant women.
The absence of a consistent translation of improved attendance on the continuum of maternal and neonatal health services were also observed in previous reviews [13][14][15]18], and may be due to the quality of the evidence with moderate risk of bias across studies, especially with nonrandomized study designs. Another factor may be substandard care provided at facilities. In fact, mHealth has been proposed as catalyst to identify those areas where strengthening is needed [40]. To explore the impact of mHealth on health outcomes in the future, studies on mHealth interventions should address health endpoints as primary or secondary outcomes and more rigorous study designs are needed that consider the gold standard components to determine an intervention's effect, i.e. a RCT either on individual or on cluster level. The need for rigorous study designs is amplified by the numerous actors involved in mHealth activities in maternal health. As the grey literature search illustrates, community based and non-governmental organizations are very active in this field. Increased attention to ensure such interventions and activities are evaluated and results disseminated will increase the evidence base, and provide solid evidence on which governments and institutions can base decisions whether to implement mHealth interventions. mHealth interventions can be implemented in isolation, at several levels within the health system simultaneously, or combined with other inter-sectorial improvements such as infrastructure and capacity of (human) resources [8,13,40,[49][50][51]. mHealth interventions that were combined with non-mHealth interventions showed positive results. Huq et al. set up a toll-free mobile communication network that allowed direct two-way communication for support and emergency situations between pregnant women and community based skilled providers (CSBA), and between CSBAs and medical specialists [52,53]. This was combined with strengthening of the health system, CSBA training, and community support group formation. Mothers perceived the communication with CSBAs as beneficial, improving access of services and overcoming issues of long distances. It also led to more prompt referrals when there were maternal complications and CSBAs could not manage. Mobile communication made it possible to inform pregnant women of the appropriate facility to seek care, decreasing delay in management. Flax et al. showed that breastfeeding learning sessions reinforced through educational text and voice messages sent to participants phones, along with songs and dramas created by participants themselves at micro-credit meetings, improved breastfeeding practices in the neonatal period (initiating breastfeeding within 1 hour of delivery (OR 2.60, 95% CI 1.60-4.10), giving only colostrum or breast milk in the first 3 days of life (OR 2.60, 95% CI 1.40-5.00)) [54]. For the transition towards comprehensive mHealth programs featuring alongside other investments, we recommend greater efforts to collaborate, bring forward and share the evidence of (mHealth) activities.
A strength of this review is that it is, to the best of our knowledge, the first comprehensive review which assesses systematically the evidence on mHealth interventions for maternal and neonatal care, and provides an evaluation of the quality. To allow for a comprehensive overview that considered the activities published outside of the peer-reviewed domain, we also included grey literature and descriptive studies. Combining all studies in a SWOT analysis provided rich data about local needs and context, the optimization in the process of mHealth intervention design and implementation, and what is needed to achieve improved outcomes. Although mHealth interventions should be adapted to local contexts [15], the SWOT analysis showed that several features are considered important across interventions. On the level of technology, the use of simple mobile phone technology and locally developed interventions facilitate implementation by building on pre-existing knowledge of mobile phone use [23][24][25]30,33]. Access is increased by providing information in lay terms, making information available in different (local) languages, maintaining low costs or toll free mobile communication, and providing voice SMS for those illiterate (Table 4) [23][24][25][26][27][28][29][30][31][34][35][36]. The latter is important, as our systematic review did find that inequities arise in who can access mHealth interventions and its potential benefits or not, depending on phone ownership, literacy, rural or urban residency and socio-economic status [23][24][25]28,32,34,35,[37][38][39][40][41][42][43]. In addition, almost all RCTs were conducted in urban areas, highlighting the importance of being aware of the possible selection mechanisms that "technology-driven tools and strategies tend to have built-in" which can affect those at the bottom of the pyramid [38].
Including the grey literature allowed us as well to observe the importance of improving collaboration between academic institutions and implementing organizations, as both can learn from each other. This may also address the incongruence between mHealth activity reports in the academic literature and those reported from the field. The quality improvement that NGOacademic collaboration can offer is illustrated by the Wired Mothers project's RCT with a low risk of bias [23][24][25]. This also enhances transparency, as the Wired Mothers project is one of the two publicly registered trials (NCT01821222) [23][24][25]. Lastly, it can result in the definition of a common language and framework when mHealth interventions and evaluations are discussed, across stakeholders and disciplines [4].
A limitation of the systematic review is that a meta-analysis was not feasible based on the data obtained from the include articles. A further limitation is the domain of this systematic review, maternal and neonatal care in LMIC, which resulted in the exclusion of several interventions of potential interest for a comprehensive mHealth intervention along the continuum of maternal, neonatal and child health [15]. These included interventions aimed at family planning [55,56], programs for immunization beyond 28 days [57,58], retinopathy of immaturity [59], and breastfeeding practices [48]. We also did not consider in this review mHealth interventions targeted at health professionals and those conducted in high-income countries. [51,[60][61][62][63][64][65][66][67]. Burden of morbidity and mortality is greatest in LMIC, and it is in these countries where mHealth is currently receiving great interest as an innovative method to contribute towards the achievement of Millennium Development Goal 4 and 5, respectively Sustainable Development Goal 3. Improved communication within the global health community, between NGOs, governments, donors, and other stakeholders, will also help to identify priorities mHealth could and should address [15], and what the position is in relation to the health system and other technological innovations in the domain of eHealth [68]. Feasibility studies and government involvement can help in identifying these priorities and aid and integration of mHealth interventions into the existing healthcare system [18], as well as guiding frameworks and cost-effectiveness assessments.
Conclusions mHealth interventions targeted at pregnant women can be an effective solution to increase service utilization to improve maternal and neonatal outcomes. The emerging trend of strong experimental research designs, feasibility research, government involvement and integration of mHealth into the healthcare system is encouraging and can pave the way to improved decision making how best to implement mHealth interventions.