Leveraging Interactive Text Messaging to Monitor and Support Maternal Health in Kenya (AI-NEO)

Mobile health (mHealth) interventions such as interactive short message service (SMS) text messaging with healthcare workers (HCWs) have been proposed as efficient, accessible additions to traditional health care in resource-limited settings. Realizing the full public health potential of mHealth for maternal health requires use of new technological tools that dynamically adapt to user needs. This study will test use of a natural language processing computer algorithm on incoming SMS messages with pregnant people and new mothers in Kenya to see if it can help to identify urgent messages.

Despite recent achievements in reducing child mortality, neonatal deaths remain high, accounting for 46% of all deaths in children under 5 worldwide. Addressing the high neonatal mortality demands efforts focused on getting proven interventions to at-risk neonates and their families. mHealth interventions have the potential to improve neonatal care and healthcare seeking by caregivers. Impact of such interventions will be maximized by ensuring healthcare workers accurately triage messages from caregivers and respond appropriately and quickly to messages that indicate an urgent medical question. This study adds to current knowledge by testing a novel natural language processing (NLP) tool to detect urgent messages. To the investigators' knowledge, such a tool has not been developed and empirically tested in a "real-world" implementation. Moreover, NLP tools to date have mostly been developed for high-resource languages; the investigators are not aware of any tools developed for detecting urgency in Swahili and Luo languages. This study's overarching hypothesis is that development of an adaptive variant of the Mobile WACh SMS platform that automatically detects and prioritizes urgent messages will be feasible and acceptable to nurses and end-users, and will reduce the time from message receipt to HCW response. Broad Objectives The study's overarching aim is to implement an NLP model into the Mobile WACh SMS platform and test its acceptability and impact on HCW response time. Aim: Pilot the adapted Mobile WACh system (AI-NEO) and evaluate its acceptability and effect on nurse response time. Eighty pregnant women will be enrolled to receive the AI-NEO SMS intervention. Women will be enrolled at >=28 weeks gestation and will receive automated SMS regarding neonatal health from enrollment until 6 weeks postpartum, and will have the ability to interactively message with study nurses. Participant messages will be automatically categorized by urgency. Intervention acceptability and recommended improvements will be evaluated among clients and nurses using quantitative and qualitative data collection at study exit (quantitative questionnaires with all client participants and qualitative interviews with 4 nurses). Nurse response time to urgent and non-urgent participant messages will be compared in the AI-NEO pilot vs. the ongoing Mobile WACh NEO trial, in which a non-adapted Mobile WACh system is used.

Participants will receive automated SMS messages with prompts to reply. They will have the ability to both respond to and initiate SMS dialogue. Trained Study Nurses will monitor and respond to participant messages. The NLP model will be applied to messages and will highlight those determined to be urgent.

This study uses Mobile WACh, a human-computer hybrid system that enables two-way SMS communication and patient tracking, to provide consistent support to women and their infants during the peripartum period and 6 weeks into the baby's life. Women will receive automated SMS messages targeting the appropriate peripartum period and will have the capability to respond and spontaneously message a nurse based at the clinic. During pregnancy, automated SMS will be delivered weekly. Two weeks prior to the participant's estimated due date (EDD), daily messaging will begin, and will continue for two weeks after delivery is ascertained. Thereafter, SMS will be delivered every other day. Women who experience pregnancy or infant loss will be enrolled into an infant loss track. The NLP model will be applied to incoming participant messages. Those flagged as urgent by the model will be flagged within the SMS system, allowing study nurses to triage and appropriately respond to those messages.

AIM (Acceptability of Intervention Measure) score (Weiner et al instrument. Score range 4-20; higher score indicates higher acceptability)

Inclusion Criteria: Pregnant ≥28 weeks gestation Daily access to a mobile phone (own or shared) on the Safaricom network Willing to receive SMS Age ≥14 years Able to read and respond to text messages in English, Kiswahili or Luo, or have someone in the household who can help Exclusion Criteria: Currently enrolled in another research study

Data from AI-NEO will be available at end of the project by contacting the study team at the University of Washington (keshet@uw.edu).