Evaluation of a Two-Way SMS mHealth Strategy to Reduce Neonatal Mortality: Rationale, Design and Methods of a Randomized Controlled Trial in Kenya

Keshet Ronen (  keshet@uw.edu ) University of Washington https://orcid.org/0000-0001-5625-4884 Esther M Choo University of Washington Brenda Wandika Kenyatta National Hospital Jenna Udren University of Washington Lusi Osborn Kenyatta National Hospital Peninah Kinthao Kenyatta National Hospital Anna Hedstrom University of Washington Millicent Masinde Kenyatta National Hospital Manasi Kumar University of Nairobi Dalton Wamalwa University of Nairobi Barbra A Richardson University of Washington John Kinuthia Kenyatta National Hospital Jennifer A Unger University of Washington


Introduction Background and rationale {6a}
Half of all deaths under the age of 5 occur in the neonatal period, de ned as the rst 28 days of life, and 75% of these neonatal deaths occur during the rst 7 days of life. 1 The proportion of under-ve mortality that occurs in the neonatal period increased from 40% in 1990 to 47% in 2018. 2 In Kenya, the neonatal mortality of 20 per 1000 births remains higher than the global average of 18 per 1000 births, and among The study is conducted at four healthcare facilities in Kenya: Mathare North Health Centre, Riruta Health Centre, Rachuonyo County Hospital and Bondo Sub-County Referral Hospital. Two facilities (Rachuonyo and Bondo) are in rural locations in Western Kenya, and two facilities (Mathare and Riruta) are in urban locations in Nairobi. These sites were chosen to include Nairobi, Homa Bay and Siaya counties, which all have high neonatal mortality. 3

Eligibility criteria {10}
Women are eligible to participate if they are ≥14 years old, pregnant at 28-36 weeks estimated gestational age, enrolled in antenatal care at a study site, have daily access to a shared or personal mobile phone with a subscriber identity module (SIM) on the Safaricom network, are willing to receive SMS, plan to stay in the area for 5 months or greater, and are not enrolled in another study. Literacy is not required if a woman has access to a partner or trusted person who she is comfortable having read her the messages. The approach of involving partners in receiving and sending SMS was developed in consultation with Kenyan mothers and successfully implemented in prior Mobile WACh studies. 51,52 Who will take informed consent? {26a} Potential participants are identi ed in antenatal clinics by clinic staff, informed about the study, and referred to study nurses for eligibility screening. Study staff obtain verbal consent from potential study participants prior to eligibility screening. If a woman is eligible for trial participation, study staff undertake the process of informed consent and obtain written consent. When potential participants are illiterate, an impartial witness is present during the entire consent process and signs the consent form. The age of consent in Kenya is 18, but pregnant women age 14 or older are considered emancipated and can consent independently, thus all eligible study participants are able to consent independently.

Additional consent provisions for collection and use of participant data and biological specimens {26b}
Consent includes permission to conduct home visits or phone interviews for tracing purposes and to conduct a verbal autopsy in case of an infant death. The consent also includes permission to use deidenti ed participant data in future studies. Biological specimens are not collected in this RCT.

Interventions Explanation for the choice of comparators {6b}
Inclusion of a control arm isolates the effect of Mobile WACh NEO on primary and secondary outcomes. All participants receive current standard of care and education provided in antenatal care (ANC) and postnatal care (PNC) by MCH clinic staff. Women randomized to the control arm do not receive SMS messages from Mobile WACh NEO. Interaction with study personnel only occurs for study data collection during follow up visits.

Intervention description {11a}
Participants randomized to the two-way SMS arm receive a series of automated SMS messages during pregnancy and until 6 weeks postpartum. Message content is designed based on estimated gestational age or time since delivery. Messages are sent in English, Dholuo or Kiswahili depending on the participant's preference. Participants can send SMS messages to the study at any time and study nurses respond by SMS during business hours within one business day through a web-based platform. Sending and receiving SMS is at no cost to the participant. SMS messaging content during the pregnancy and postpartum periods are summarized in Figure 2. From enrollment up to 38 weeks gestation, automated SMS are delivered weekly, encouraging facility delivery and birth planning, and providing anticipatory information about neonatal danger signs and emotional support. Automated SMS are delivered at a pre-speci ed day, time and language based on participant preferences. From 38 weeks gestation until delivery, women receive daily messages highlighting identi cation of neonatal danger signs and ENC practices including immediate and exclusive breastfeeding, thermal and cord care. From delivery to 2 weeks postpartum, mothers receive two messages per day: one message with screening questions for identi cation of infant danger signs and encouragement to engage with nurses by SMS if they have a concern, and one message with educational content on ENC practices, mental health and support. From 2 to 6 weeks postpartum, SMS are delivered every other day. All messages are personalized with the participant's preferred name and contain a prompt to engage women in message content. Participants in both the control and intervention arms receive standard ANC and MCH clinical services through MOH clinics with minimal in-person interactions with study personnel.
Messages were adapted from prior Mobile WACh studies, 51-53 consistent with Kenyan Ministry of Health guidelines for newborn care, 54 and tested during a pilot study involving 800 peripartum women in Mathare North Health Center and Rachuonyo County Hospital. 46 The messaging curriculum was modi ed through consultations with obstetrics, neonatology, and mental health specialists and literature review on supportive messaging. Messages are personalized by opening with the participant's name. They are action-oriented and address a speci c outcome of the trial. Additional messaging "tracks" were created with adaptations to the messages for mother-infant dyads at elevated risk for obstetrical and neonatal complications. Those placed in high-risk tracks receive additional messages tailored to their speci c risk factor. High-risk groups include (1) rst-time mothers or mothers age ≤19, (2) women who screen positive for elevated depression symptoms during study data collection using the Edinburgh Postnatal Depression Scale (score ≥13) or (3) women whose infants are born premature (<37 weeks estimated gestational age) or are of low birth weight (<2.5kg). Additionally, if a participant loses her pregnancy or her infant dies and she consents to continued messaging, she is switched to a message track that contains no content related to pregnancy or infant care but includes emotionally supportive condolence messages designed in a previous Mobile WACh study. 55 Messages sent to participants in speci c tracks attempt to capture a broad group at higher risk of poor pregnancy or neonatal outcomes who may bene t from targeted messaging.
The Mobile WACh NEO SMS curriculum was developed based on the IMB theory, which holds that health behavior is in uenced by an individual's knowledge of information about the bene ts of the behavior, their motivation to change behavior, and their skills to change and engage in new behavior ( Figure 1). 50 Based on this model, topics and example messages sent in the study are summarized in Table 1. A custom human-computer hybrid software system is used to manage incoming and outgoing SMS for Mobile WACh NEO ( Figure 3). 52,56 This system enables seamless two-way SMS communication and participant tracking. To facilitate management of messages, the messaging platform was developed through a human-centered design processes with Kenyan nurses to improve usability and streamline work ow. 56 The interface enables nurses to review incoming participant SMS and counsel on potential medical conditions or concerns within one business day. The system also has a bank of responses to frequently asked questions, such as common pregnancy concerns, birth planning, infant feeding, and infant illness, which nurses can use as a modi able template for their responses. This allows for improved standardization and response times. Nurses also can freely respond to participant messages or questions that arise. Nurses triage messages as they are received and respond with greater urgency to messages involving a reported maternal or neonatal danger sign, serious adverse event, or home delivery. If participants report neonatal or maternal danger signs, they are referred to facility care by study nurses.
If a participant reports experiencing intimate partner violence, suicidal ideation, elevated depressive symptoms, or severe food insecurity, study staff refer the participant to available local resources. Prior to study initiation, study personnel received training on use of the SMS system and topics in neonatal, maternal health and psychosocial support. Study messages are reviewed weekly by senior team members and discussed with study nurses for quality assurance.

Criteria for discontinuing or modifying allocated interventions {11b}
Participants enrolled in the intervention arm can opt out of the intervention at any time or send an SMS to withdraw from the intervention. If a participant reports a pregnancy or infant loss, they automatically stop receiving automated messages and are offered the option to continue receiving modi ed messages, as described above. To comply with movement restrictions during the COVID-19 pandemic, home-tracing is not conducted.

Strategies to improve adherence to interventions {11c}
All SMS messages include a question to the participant to encourage interaction with study nurses. To ensure timely transition from antepartum to postpartum messaging at delivery, participants are encouraged to report their delivery as soon as possible by SMS or phone call. Participants without a recorded delivery by 1 week after their estimated delivery date are contacted by phone call to con rm their delivery status. If they are not reached by phone tracing, home tracing is performed. If delivery is not con rmed by 43 weeks gestational age and tracing is unsuccessful, the participant automatically starts receiving postpartum messaging.

Relevant concomitant care permitted or prohibited during the trial {11d}
Study participants receive all routine clinical care at the facility. All participants who are identi ed as experiencing intimate partner violence, suicidal ideation, elevated depressive symptoms, or severe food insecurity are referred to available resources and additional assistance regardless of study arm.
Provisions for post-trial care {30} No provisions for post-trial care are offered to participants.

Outcomes {12}
The primary outcome of the study is neonatal death, de ned as death within the rst 28 days after live birth. All study outcomes are outlined in Table 2. The participant timeline is depicted in Figure 4. Participants enter the study at 28-36 weeks gestation and are randomized to either the Mobile WACh NEO or control arm at enrollment. At 2 and 6 weeks following delivery, participants attend a study visit in conjunction with their usual medical care at the clinic or complete a study visit by phone, in which study nurses verbally administer a questionnaire to ascertain delivery information and data on primary and secondary outcomes. Following the 6-week study visit, indepth interviews are conducted with 60 intervention participants to gather data on their experience with the SMS curriculum. In addition, verbal autopsies are conducted via phone or in person 6 weeks or more after an infant death occurs. Participants are enrolled in the study until 18 weeks after delivery to allow time for all follow-up activities, including in-depth interviews and verbal autopsies as needed.

Sample size {14}
The study is powered to detect a difference in the primary outcome of neonatal mortality. With 5,000 participants randomized 1:1 to SMS and control, assuming 10% attrition and alpha=0.05, we have 80% power to detect a hazard ratio of £0.53 assuming a neonatal mortality rate of 23 per 1000 in the control arm. With this sample size, assuming alpha=0.0045 (Bonferroni-adjusted for 11 tests, see data analysis section below), we will also have 80% power to detect the following changes in secondary outcomes: a risk ratio of 1.05 in early initiation of breastfeeding assuming 80% uptake in controls, a risk ratio of 1.11 in application of substances to the cord or delaying rst bath assuming 50% in controls, a risk ratio of 0.76 in elevated depression symptoms assuming 19% in controls.
Sample size for post-RCT qualitative data collection was determined based on the number of interviews needed to achieve saturation of concepts. It is estimated that 60 interviews will be su cient to achieve saturation and provide relevant intervention and implementation feedback.

Recruitment {15}
Study staff obtain a list of women presenting for ANC visits at the clinic each day. They then approach potential participants at ANC visits and provide a short description of the study. If a potential participant is interested in learning more about the study, study staff perform eligibility screening following the antenatal care visit.

Sequence generation {16a} and Concealment mechanism {16b}
Enrolled pregnant women are randomized to two-way SMS or control, using 1:1 allocation (Figure 4). A web-based randomization service generates allocations, strati ed by site, using random block sizes of 2, 4, and 6 (Randomize.net, Ottawa, Canada). The randomization service only displays and records the allocation after staff have entered the enrolled participant ID. The study arm is unblinded to participants and study staff. To ensure participants are distributed across sites, no site will enroll more than 2499 of the 5000 total women. In rare situations where randomization using the web-based service is not possible due to technical issues, envelopes containing randomization allocations generated by the web-based service are available.

Implementation {16c}
The allocation sequence is generated by the web-based randomization service and is not available to investigators or staff in advance of randomization. Study staff at each of the four sites enroll and assign participants to the arms from the randomization allocations obtained through the web-based randomization service.

Assignment of interventions: Blinding
Who will be blinded {17a} Participants and study staff are not blinded. The principal investigator and co-investigators are blinded to study ndings until study close.

Procedure for unblinding if needed {17b}
There is no provision for unblinding as both participants and study staff are unblinded in this trial.

Data collection and management
Plans for assessment and collection of outcomes {18a} Data are collected at study visits at enrollment in pregnancy, 2 weeks postpartum, and 6 weeks postpartum. At enrollment, a tablet-based enrollment questionnaire using an open-source tablet-based data collection system (KoboToolbox) is used to collect data on demographics, clinical and pregnancy history, family planning, experience with SMS and technology, social support, intimate partner violence, maternal and child health status, breastfeeding plans, parental self-e cacy, and depression. To determine estimated delivery date (EDD) at enrollment, women are asked the date of the rst day of their last menstrual period. EDD is also abstracted from the MCH booklet, the primary medical record used in the peripartum period, when available. pandemic, the research team modi ed study procedures to suspend any in-person home visits or clinic visits only for data collection procedures. Follow-up data collection occurs via in-person study visits aligned with routine clinical care or via phone, following Ministry of Health guidelines and recommendations to reduce possible exposures to COVID-19. The study also complements data collection at study visits with abstraction of clinic record data about hospitalizations and deliveries. The primary source of abstracted data for maternal or infant hospitalization is the hospital discharge form brought by the mother to the 2-or 6-week study visits. Medical record abstraction is also performed to collect delivery details if a participant misses the 2-or 6-week study visit(s). The primary source of abstracted data on delivery is the MCH booklet brought by the mother and the maternity registry at the clinic. Abstracted data are entered into an electronic data collection system using REDCap.
When the study is informed of an infant death, a verbal autopsy is conducted to determine the primary causes of death, using the Population Health Metrics Research Consortium (PHMRC) shortened verbal autopsy neonatal questionnaire. 57 The primary cause(s) of death are generated using the SmartVA -Analyze 2.0.0 application.
To ensure data quality, questionnaires at each study visit contain programmed range checks for plausible data values and automatic skip patterns. Data checks are performed on a weekly basis to identify data errors.

Plans to promote participant retention and complete follow-up {18b}
Active tracing by phone is conducted if a study participant misses a 2-or 6-week study visit by 1 week or no delivery date is recorded by 1 week after EDD. If the participant is not reached by phone after four attempts, home tracing is conducted. Participants who do not wish to come to clinic for the 2-or 6-week study visit are offered to complete the questionnaire by phone or at home if suitable locations for ensuring participant con dentiality are available. To comply with movement restrictions during the COVID-19 pandemic, the follow-up visits, verbal autopsy and in-depth interviews are offered via phone and home-tracing is not conducted.

Data management {19}
Data are collected digitally and uploaded daily to a secure cloud storage location accessible only by key data management personnel. Access to electronic data capture systems used in this study, including KoboToolbox and REDCap, is limited to study staff who conduct data entry and management. Data systems, including the SMS platform, are password protected and accessible only by authorized users. The data manager generates weekly reports for study monitoring and to identify errors in data. Resolution of data errors is documented thoroughly in data management logs and during data cleaning.

Con dentiality {27}
Any personally identi able information related to participants is maintained separately from study data to preserve participant con dentiality. Paper forms such as signed consent forms are stored in locked cabinets separate from forms connecting participant names and their study participant identi er. All participants are assigned a study identi cation number (ID) and identi ed only by that ID after enrollment. Discussions between study staff and potential participants during study visits or for purposes of data collection are conducted in a private space to maintain privacy.
Plans for collection, laboratory evaluation and storage of biological specimens for genetic or molecular analysis in this trial/future use {33} There are no plans for collection of biological specimens.

Statistical methods for primary and secondary outcomes {20a}
Study outcomes and analytic approaches are summarized in Table 2. The primary study outcome is neonatal mortality, de ned as death within the rst 28 days of life as reported in clinic records or maternal report. Risk of neonatal mortality will be compared between arms by log-binomial regression. All analyses will be intent-to-treat. Analysis of secondary outcomes will be corrected for multiple comparisons using the Benjamini-Hochberg method. 58 Power calculations were conducted using a Bonferroni adjustment to provide a conservative estimate of the limits of detectable differences.
Methods in analysis to handle protocol non-adherence and any statistical methods to handle missing data {20c} Primary analysis will be intent-to-treat, but an exploratory per-protocol analysis will be conducted excluding women who discontinue the SMS intervention. Based on previous studies in this population, we expect 10% loss to follow-up between enrollment and the 6-week study visit. Sensitivity analyses will be conducted for the primary mortality outcome assuming all participants who are lost to follow-up and missing mortality data did or did not experience neonatal mortality.

Interim analyses {21b}
An interim analysis for neonatal mortality will be performed using O'Brien-Fleming boundaries for bene t and harm when 50% of expected person time has been accrued.
Methods for additional analyses (e.g. subgroup analyses) {20b} Participants from the intervention arm will be purposively sampled to participate in IDIs based on their engagement in the Mobile WACh NEO intervention. Participants will be divided into high, medium and low interactor groups based on messaging behavior collected on the Mobile WACh system. From each group, ten participants will be randomly selected to complete interviews. These interviews will focus on barriers and facilitators of SMS engagement to characterize high and low users. In addition, we will interview thirty women whose neonates had complications or died to understand perceived utility of remote engagement with nurses during infant illness.
Plans to give access to the full protocol, participant level-data and statistical code {31c} The full protocol is provided as a supplement. Participant-level data on primary outcomes and statistical code will be made available after publication of trial ndings, upon request from the authors.

Oversight and monitoring
Composition of the coordinating center and trial steering committee {5d} The trial study team, consisting of the PI, site-PI, research coordinators, data manager, pediatrician, neonatologist, epidemiologist, psychologist, Kenyan MOH consultant and research assistant, meet on a weekly basis to provide oversight of the study. The University of Washington serves as the trial coordinating center, providing overall study leadership, data management, ethical compliance, budget management and oversight of study operations. The data management team consists of the data manager, epidemiologist, research coordinator, study coordinator and research assistant who oversee data quality and meet on a weekly basis.
Composition of the data monitoring committee, its role and reporting structure {21a} The All changes to the study protocol are submitted for review by the ethics review committees at the University of Washington and Kenyatta National Hospital/University of Nairobi. The DSMB and clinicaltrials.gov record is updated with any updates to data collection or analysis.

Dissemination plans {31a}
Dissemination of study results to health researchers, policymakers, and clinicians has potential to improve prevention strategies for neonatal mortality in Kenya and in other countries. All data collected in this study will be publicly disseminated to key policymakers at county and national level in Kenya, the public health research community in Kenya and through peer reviewed international journals and conferences. Results from the study will be reported no later than one year after the completion date. The study has been registered in ClinicalTrials.gov.

Discussion
The Mobile WACh NEO trial will evaluate the impact of a peripartum two-way SMS intervention on neonatal mortality. The intervention is designed to impact neonatal survival through improved ENC practices, neonatal care-seeking behaviors, and maternal mental health, which will also be measured as secondary outcomes.
SMS has the potential to reach marginalized populations at scale given the expansion of mobile service coverage and widespread access to personal cellular phones. A growing number of programs leverage mobile infrastructure to deliver health information and support in the peripartum period, including several that have been nationally scaled, illustrating their potential public health impact. 59 This addresses critiques that mobile health studies lack emphasis on identifying modi able components in the pathway that impacts neonatal outcomes. 69 Inclusion of self-e cacy, social support, and depression as intermediate outcomes undergirds the importance of approaching neonatal interventions in a more holistic approach to determine the impact of maternal psychosocial health on neonatal outcomes.
The Mobile WACh NEO intervention is also designed to be practical in resource-limited settings. The platform facilitates both sending of automated messages to participants and interactive messaging with a nurse. Scheduled messages are sent to participants in bulk, while replies and individually initiated messages are read and responded to by a nurse, creating an e cient system to maximize healthcare worker time in triaging messages. In addition, automated messages are tailored to women's risk factors and current stage of pregnancy with more frequent messaging at times when risk of adverse neonatal outcomes is highest. Finally, this study is conducted in a mix of rural and urban clinics in Kenya to increase generalizability across geographical contexts.
This study has some limitations. While the criteria used to enroll participants into the high-risk tracks are targeted, messages may not always be fully applicable to all participants in every circumstance. Due to feasibility considerations in implementing the intervention, participants are ineligible if they lack daily access to a Safaricom phone. Phone access in Kenya is now 89%, and Safaricom is the most widely used mobile phone provider in Kenya, but excluding women who lack access to a Safaricom phone may introduce bias. 35,36 Also, if a participant is unable to read and write SMS messages, their participation in two-way SMS is reliant on an individual who can read the SMS messages to them, so engagement by low-literacy individuals may be more limited. This may reduce generalizability to low-literacy women or those who do not have an individual who can assist them, who may be the women most in need of such an intervention.
Mobile WACh NEO is unique in its approach to evaluate the effect of a semi-automated two-way SMS system on neonatal outcomes, integrate maternal behavioral health with neonatal health, and understand the underlying mechanisms. Rigorously evaluated interventions will provide a stronger evidence base for decisions about scaling mobile health technologies and maximize their impact.