Analytics & Machine-learning for Maternal-health Interventions (AMMI): A Cross-CTSA Collaboration - PROJECT SUMMARY African-American women across the US experience alarmingly higher rates of maternal mortality than their white counterparts. Factors associated with social determinants of health (SDoH), including education, housing, transportation, and nutrition are recognized as potentially contributiing to this disparity in maternal health outcomes, along with clinical risk factors including hypertension and heart disease. However, the complex associations among these factors, along with the causal role they play in increased risk for maternal mortality, are not well understood, nor are there comprehensive health care interventions that take these combined factors into account to provide decision and communication support for patients, providers, and community support workers. The Analytics and Machine-learning for Maternal-health Interventions (AMMI) initiative, a collaborative effort from researchers at UNC- Chapel Hill, Duke, and Wake Forest, aims to address these gaps by developing a machine learning- enhanced health technology framework to reduce downstream risk of maternal mortality in African- American women. By integrating data across the three institutions that includes both clinical and SDoH factors, and by building machine learning applications grounded in this data, AMMI’s goals are to: 1) clarify and track contributions of biological, clinical, and SDoH factors toward specific maternal morbidities associated with eventual mortality, 2) conduct efficient and accurate risk predictions to determine whether patients fall into defined target risk groups, and 3) translate these risk predictions into interventions appropriate for providers, patients, and community support organizations. A key focus of the initiative is to create an advanced technology infrastructure supporting connectivity and communication among these three types of stakeholders, with the goal of building trust and awareness based on automatically curated decision support aids and ultimately mitigating patient risk. To this end, Aim 1, focused on establishing system requirements, begins with the formation of a stakeholder group that brings together patient, provider, and community support organization representatives to engage in design and evaluation with AMMI researchers throughout the project. Aim 2 focuses on systems development, including the creation of 1) a custom-built clinical and SDoH data mart, 2) clinical decision support software using machine learning algorithms, and 3) three user-facing apps aimed at providers, patients and community support personnel, and AMMI researchers. Aim 3 focuses on pilot-level deployment of the system, integrating the AMMI apps through Epic to provide informational interventions to providers, patients, and community support personnel. Aim 4 engages stakeholders in formative and summative evaluation during and after the deployment phase (Aim 3), including both testing of the software function and measurement of the impact of AMMI interventions on end users.
