AI/ML to Reduce Maternal and Infant Health Disparities
Summary
In 2023, MedStar Health D.C. Safe Babies Safe Moms (SBSM) committed to building a model electronic health record (EHR) surveillance system to improve safety for birthing individuals and their babies. This innovative surveillance system will use artificial intelligence and machine learning to systematically identify negative tone, negative sentiment, and potentially stigmatizing language in EHR, alongside signals of risk factors for adverse birth outcomes, as well as signals ensuring proper care delivery. Findings will be used to improve health service provision and ensure that birthing individuals and their babies have the resources needed to meet their full health potential.
Approach
MedStar Health commits to expanding its existing Safe Babies Safe Moms (SBSM) initiative to strengthen identification of risk factors for adverse maternal and infant outcomes. MedStar Health will develop a new surveillance system that monitors the electronic health record to identify signals of these risk factors – an expansion of SBSM that aligns with its goal of enhancing technology to improve outcomes. AI/ML will be used to identify clinical patterns of care and translate them into informatics-based queries, resulting in safer and more equitable systems and processes for birthing individuals and their babies.
Results of SBSM’s AI/ML research will be used to build the system. In-house and contracted team members will develop surveillance techniques to systematically identify negative tone and sentiment, potentially stigmatizing language, those who have clinical diagnoses that are at risk for adverse birth outcomes, and proper treatment plans for those at risk. The continuing use of AI/ML will allow SBSM to prospectively identify patient cohorts at risk and use this data to continue to improve the system.
The new system will directly impact patients by improving care and safety for birthing individuals and their babies. For example, Black women are about three times more likely than White women to die at any point of pregnancy, and late maternal deaths are 3.5 times more likely among Black women than White women. A major cause of complications and death is postpartum hemorrhage. The SBSM system would identify patients at risk for hemorrhage before delivery, including those with anemia, and monitor to ensure proper treatment and the appropriate use of tone and sentiment. The system will provide the appropriate feedback to care teams for necessary adjustments in care plans, as well as potentially biased care. This system will not only ensure the care teams provide medically appropriate care, but also that it is delivered in an unbiased and nondiscriminatory way.
Action Plan
The goal is to use advanced techniques in artificial intelligence, machine learning, and clinical informatics to develop a quality improvement system that can be used for surveillance to improve patient care and safety. The final product will be a surveillance system implementation guide, which can be shared with other systems to adapt according to their needs. Throughout the project’s duration, the team will use data and lessons learned to continuously improve patient care and safety, which will result in positive impacts on some patients even before the final product is completed. Product development will include three “cycles” or product development periods. In year one, Medstar Health will focus on identifying patterns in the surveillance data. In year two, they will pilot interventions based on learning from year one, clean up language processing, and begin allowing MedStar Health clinicians and researchers to use data generated from this work to engage in relevant research. In year three, Medstar Health expects to have a minimum viable product (MVP) , analyze the effectiveness of our interventions, and complete the final report.
2024
Q1: kick off the project and complete overall design framework.
Q2: Launch Cycle 1, begin pattern detection analysis.
Q3 and Q4: Continue designing and building the system based on analysis.
2025
Q1: Launch Cycle 2 and begin research using data generated.
Q2 and Q3: Continue adding features and fixing issues.
Q4: Launch Cycle 3.
2026
Q1: Continue with Cycle 3.
Q2: Complete analysis of Cycle 3.
Q3 and Q4: Complete final analysis and report, including application design specifications.
Background
Pre-term birth and maternal and infant mortality and morbidity rates in the District of Columbia are among the highest in the nation: The CDC reports that the maternal mortality rate in the District is more than twice that of Los Angeles and 50% greater than New York City. Black families in the District account for 50% of all births and a staggering 90% of all pregnancy-related deaths (Maternal Mortality Review Committee Annual Report, 2021) . Where a person lives also plays a major role, with 63% of all pregnancy-related deaths occurring east of the Anacostia River in Black-majority Wards 7 and 8. Additionally, infant mortality rates in Wards 7 and 8 are twice the District average (DC Kids Count) .
The CDC reports that more than 80% of pregnancy-related deaths were preventable and highlights the need to use data to prioritize interventions that save lives and improve health disparities. To prevent deaths, maternal health researchers and providers must first understand the causes of severe maternal morbidity – nearly 90% of which were preventable. For every mother who dies from childbirth, 70 mothers die and suffer from severe maternal morbidity. But even focusing on severe morbidity is too narrow; maternal safety should emulate the approach of aviation safety and some patient safety experts by not only investigating patient harm events resulting in serious injury or death, but also conducting surveillance across a full spectrum of patient safety. These efforts will help uncover unsafe conditions, hazards, near-misses, and other maternal injuries, not resulting in serious morbidity or mortality.
Research indicates that artificial intelligence (AI) and machine learning (ML) coupled with electronic health records (EHR) can be useful for identifying at-risk birthing individuals and identifying patterns associated with poor health outcomes. DC Safe Babies Safe Moms (SBSM) leaders are using AI as one strategy to improve maternal and infant health outcomes and reduce racial disparities.
Progress Update
The AI/ML to Reduce Maternal and Infant Health Disparities Commitment has produced two functional dashboards and a working machine learning model.
Using natural language processing techniques, the informatics team developed a machine learning model to review providers’ clinical notes in the electronic health record (EHR) for instances of negative tone and sentiment, allowing leadership to employ early intervention and targeted training. A newly completed pipeline sends EHR data to the Databricks Data Intelligence Platform, allowing the Safe Babies Safe Moms (SBSM) team to process and analyze system-wide data. The team developed a tone and sentiment dashboard to share the model’s findings with hospital leadership; refinement of the dashboard’s data visualization tools will facilitate leadership’s efforts to detect and monitor providers who demonstrate negative tone. The team will incorporate user feedback to refine the tone and sentiment model and analyze site- and department-specific trends.
In collaboration with subject matter experts, the SBSM team created the Maternal Hypertension Dashboard, which provides clinical leaders with patient-level, interactive data visualization for prenatal hypertension management. Updated weekly with EHR data, the dashboard facilitates ongoing risk assessment and allows clinical leaders to monitor trends, review charts, and identify pregnant patients eligible to enroll in the specialized hypertension care bundle. Future efforts will focus on developing a model and dashboard to better detect maternal diabetes and related conditions. This dashboard will monitor patient- and system-level trends, provide weekly risk assessments, and track patients enrolled in the diabetes-specific care bundle.
The team continues to leverage common data models and international data standards (Observational Medical Outcomes Partnership) to facilitate widespread, cost-effective dissemination and adoption of developed algorithms.
Note that the secured funding amount in this report accurately reflects monetary awards specific to this commitment, as in-kind gifts from a partner organization included in the previous report were not realized.
Partnership Opportunities
The use of artificial intelligence and machine learning to improve public health is a relatively new field. Medstar Health welcomes financial resources and partners with experience in this area as they work to develop a revolutionary approach to improving maternal and infant health and reducing racial disparities.,Medstar Health’s goal is to build a model surveillance system that ultimately could be used by other healthcare organizations in the District of Columbia, other parts of the U.S., and throughout the world to improve safety for birthing individuals and their babies. MedStar welcomes partnership with other organizations pursuing similar goals and will disseminate research emanating from their CTA.