Rachel Harvill: The future of equity in maternal and child health research

As a Wallace Center Trainee Award recipient, I attended Stanford’s Maternal and Child Health Research Institute Symposium in October 2021. This conference highlighted the latest research on the use of data-driven technology to improve the early identification of risks associated with pregnancy and childbirth. As a computer scientist pursuing an MPH, it sparked my interest in how artificial intelligence might improve equity in maternal and child health.

Dr. Nima Aghaeepour is an Assistant Professor of Anesthesiology, Perioperative, and Pain Medicine at Stanford University and an active investigator at the March of Dimes’ Prematurity Research Center. Current clinical practice relies on using a variety of diagnostic tests and procedures, such as pelvic exams and ultrasounds, to identify patients at risk for preterm birth. Rather than developing a diagnostic test for one condition at a time, Dr. Aghaeepour’s research aims to use multiple data sources to develop a holistic understanding of pregnancy and early life. Dr. Aghaeepour’s team developed a machine learning model that accurately predicts preterm birth using immune system data gathered from urine samples – providing an opportunity for streamlined identification of patients at risk for preterm birth.

Through their research, Dr. Aghaeepour’s team learned a limitation of using standard machine learning algorithms for maternal and child health research questions — a single machine learning algorithm cannot be applied to different locations due to environmental and social factors that impact preterm birth. Given my previous experience as a Healthcare Data Engineer, I know that this problem is not uncommon in healthcare data analytics. I am interested in seeing how advances in algorithm development address this limitation.

Dr. Aghaeepour’s team also found that stress-based metrics gathered from stress questionnaires can predict multiple adverse pregnancy outcomes. Interestingly, these stress-based metrics can also be used to predict immune characteristics. These findings provide an opportunity for stress-based questionnaires to be incorporated into machine learning models. Listening to this portion of the talk made me think about toxic stress, the prolonged activation of stress response systems often due to social stressors such as racism, and its impact on maternal and neonatal outcomes. I would be curious to learn how these models could be adapted to account for toxic stress.

In a separate talk, Dr. Suzan Carmichael discussed how combining severe maternal morbidity, sociodemographic, and social determinants data can help identify trends in the relationship between social-structural determinants of health and severe maternal morbidity. Similar to how various data sources are needed to predict preterm birth, investigating severe maternal morbidity involves linking hospital discharge data, birth and fetal death certificates, geocoded addresses, and vital records data. Due to the lack of qualitative data, existing severe maternal morbidity research doesn’t capture upstream social determinants of health. These talks highlighted the opportunity for data-driven technology to improve equity in maternal and child health. Artificial intelligence technology has the potential to reduce the number of diagnostic tests and procedures needed to identify patients at risk for adverse pregnancy outcomes, thereby reducing costs and improving the accessibility of care. Moreover, population-based data linkage technology allows equity to be centered in severe maternal morbidity research. RAND Corporation utilized both technologies in their research on using an innovative database to build machine learning models that predict infant mortality risk and identify potential interventions. Going forward, I’m curious to see how research advances to incorporate qualitative, community-centric data in models to further progress equity.