Community demonstrations of Rosie elicited high interest from new mothers and the organizations that serve them, highlighting the need for resources to provide new mothers with easy access to reliable health information.

Maternal and infant health is an area where health disparities are particularly evident within communities of color. Additionally, mothers have high information needs during pregnancy and the first year of life. To address this, our team is developing a free, AI-powered chatbot, Rosie, capable of answering questions on pregnancy, postpartum, and infant care.

Chatbots have gained significant popularity due to their scalability and success in individualizing resources. In recent years, scientific communities and researchers have started recognizing this technology’s potential to inform communities, promote health outcomes, and address health disparities. Potential advantages of using chatbots as an intervention include a reduction in perceived judgment by allowing users to ask as many questions as needed about topics that may otherwise be considered inappropriate or stigmatized. Additionally, chatbots can help inform decision making for pregnant and newly parenting mothers by helping address concerns related to stress, sleep, and breastfeeding. Chatbots can also positively affect childhood development by providing detailed information on infant developmental stages, immunizations, health screenings, rashes and fevers, and nutrition tips among other popular topics.

The development of Rosie is an interdisciplinary project with teams focused on the technical build of the mobile app, machine learning models, and community outreach.

To develop a robust knowledge base for Rosie, a list of verifiable web sources of information on maternal health and infant care were generated. Sources included government agencies, professional medical organizations, and children’s hospitals (n=60). A corpus of documents on maternal and infant health was built by scraping text from these vetted web domains. Each web document was then parsed into passages (n=73,000) by splitting the text according to a collection of heuristics that retain sentence context.

Rosie’s end-to-end system architecture consists of a user-facing mobile application (“client”), a backend server, and a question-answering system. After a user inputs a question, the backend server receives and processes it before directing it to the question-answering system where natural language processing systems understand the query and retrieve the relevant information from the knowledge base. The system then returns a response via the client to the user. The underlying question-answering (QA) system uses an unsupervised dense passage retrieval model, Contriever, designed to find the document that best answers a user question. The model first indexes documents by pre-computing and storing their vector representations. When a user poses a question to the system, a vector representation of the question and find the document in the corpus that has a vector representation most like the question vector is computed. Along with the passage, users are shown the URL and provided a link to the web sources that the passage was extracted from to further contextualize our system’s answer.

We utilized community-engaged strategies to test the current iteration of the chatbot app. The research team reached out to community-based organizations such as farmers markets and city-wide festivals as well as organizations focused on supporting maternal and child health such as diaper banks and baby centers. From June to October 2022, we organized over 20 demonstration sessions conducted in Washington, DC; Maryland; and Virginia. On-site demonstrations were led by two or three research team members who would approach pregnant or young mothers to demo the app. Eligible participants were female, 14 years and older, currently pregnant or have a child under the age of three.

Read our article in JPHMP

A total of 109 pregnant women and new mothers of color participated in the community demo events. Among the participants, about 75% said they searched online for maternal and infant health information at least weekly. Some participants verbally expressed their discontent with using internet-based searches because of concerns regarding the validity of the information as well as the reliability of the sources. They were heartened to learn that Rosie derives responses to questions solely from vetted government and medical sources. Over 90% of participants expressed a likelihood to use Rosie. When given the opportunity to ask Rosie questions, most participants asked questions about their baby’s health and care (generally baby’s development and nutrition), accounting for about 80% of the total questions asked. Mother-related questions were mainly about pregnancy, such as labor, abortion, and mental health.

The high level of interest in the chatbot is a clear indication of the need for more resources. Rosie aims to help close the racial gap in maternal and infant health disparities by providing new mothers with easy access to reliable health information. Our community-driven application development approach ensures that Rosie is tailored to the needs of the target audience.

To learn more, read our article “Practical Guidance for the Development of Rosie, a Health Education Question-and-Answer Chatbot for New Mothers” in the new issue of the Journal of Public Health Management and Practice.

Heran Mane is a Data Analyst in the Epidemiology and Biostatistics department in the University of Maryland School of Public Health. Heran holds a BS in Psychophysiology from Lindsey Wilson College and a certificate in Data Analysis and Visualization at George Washington University. She applies her skills in research that targets public health biases, health inequities and social media research. She is also contributing to the development of a question-answering chatbot to serve mothers and pregnant women of color.

Elizabeth Aparicio, PhD, MSW, is an Associate Professor in the Behavioral and Community Health department in the University of Maryland School of Public Health. She directs the Community THRIVES Lab, a research group that conducts community-engaged transformative health research at the intersection of family violence, early childhood, and adolescent sexual health intervention. She is also the deputy director for clinical training and intervention for the University of Maryland Prevention Research Center. She is currently co-Principal Investigator of a R01 grant with Dr. Nguyen to build a question-and-answer chatbot for racial/ethnic minority mothers.

Quynh Nguyen, PhD, MSPH, is an Associate Professor within the Department of Epidemiology and Biostatistics at UMD. Her research focuses on leveraging data science to study and intervene on health disparities. She has leveraged geotagged Twitter data to characterize the social environment (K01ES025433; K01ES025433-03S1) and leveraged Google Street View data to characterize the built environment (R01LM012849). She is currently co-Principal Investigator of a R01 grant with Dr. Aparicio to build a question-and-answer chatbot for racial/ethnic minority mothers.