Find care now
If you are experiencing a medical emergency, please call 911 or seek care at an emergency room.
MedStar Health investigators study how applying algorithms to electronic health records can find red flags in a pregnant patient’s medical information that may put them at risk for maternal health complications.
Despite myriad medical advances, maternal mortality is still a challenge in the U.S. It is estimated that 70% of maternal deaths and life-threating complications are preventable.
According to the Centers for Disease Control & Prevention, the maternal mortality rate was 23.8 deaths per 100,000 live births in 2020, compared with 20.1 in 2019. For Black women, the rate was 2.9 times higher than among white women.
As part of our system-wide initiative to improve maternal and infant health care, MedStar Health Research Institute has embarked on a clinical study to determine how machine learning can help identify patients at risk of cardiovascular complications in pregnancy—and potentially prevent more maternal deaths. This research was published in JMIR Publications earlier this year.
Machine learning is a type of artificial intelligence that specializes in data interpretation, using algorithms to simulate human learning and incrementally improving accuracy. Originating from concepts devised by the mathematician Alan Turing in 1935, today’s technology lets the data talk to us, revealing trends and allowing us to analyze data from multiple sources simultaneously—such as disparate electronic medical records (EMRs).
Using an adapted version of the machine learning tool, Health Outcomes for all Pregnancy Experiences–Cardiovascular-Risk Assessment Technology (HOPE-CAT) by Invaryant, our joint research team analyzed a large set of retrospective EMR data from patients who were pregnant, some of whom developed cardiovascular conditions.
We confirmed that machine learning models can effectively identify cardiovascular risk in pregnant patients early enough to potentially avoid complications. For example, we could determine which patients were at risk of preeclampsia—life-threatening high blood pressure in pregnancy—up to 60 days before they were diagnosed.
This type of crucial information, delivered in an easy-to-access format, could provide timely opportunities for clinicians to open a dialogue about how patients can manage their heart health during and after pregnancy.
Empowerment in standardized data.
When patients seek care from different health centers, their EMR data can become fragmented. Unstandardized data, such as free-form text fields, are difficult to assess for manual risk stratification, reducing the ability to easily identify red flags with the patients’ health. These issues are especially prevalent in maternal health care, as pregnant patients receive care from providers in clinic and hospital settings that may not be related or connected, exacerbating maternal morbidity and mortality.
By extracting, standardizing, and analyzing retrospective data from 32,409 patients in multiple Cerner EMRs using an adapted version of the HOPE-CAT tool, we were able to identify early signs of risk such as elevated blood pressure, shortness of breath, and chest pain among patients who ultimately developed new or worsening cardiovascular conditions during pregnancy.
Collated into an artificial neural network, trends identified by the tool could alert providers to real-time risk factors early enough to intervene and improve a patient’s chances for positive outcomes or even prevent more serious health issues during and after pregnancy.
All the while, HOPE-CAT will sift through a growing body of diverse data, improving its predictive capabilities and identifying previously unrecognized risk-related patterns clinicians can act upon. And with the next phase of our research, we are working to make these opportunities even easier for clinicians to access.
What's next? Natural language processing.
MedStar Health Research Institute is working to include additional types of hard-to-quantify data into the HOPE-CAT tool. Using natural language processing (NLP) technology, we can extract and analyze actionable data from unstructured formats, such as provider notes.
While more research is needed, implementing NLP-driven risk stratification at the clinical level could be a major step toward health equity, providing amalgamated insights into- factors like access to food, housing, and other social determinants of health that often are captured in these free-text fields.
There is great power in data. Though machine learning will never replace a provider’s clinical empathy and expertise, it can ease the burden of data analysis at the point of care, affording patients more quality face time with providers.
And with providers and technology working together to identify potential risk factors earlier in pregnancy, we are one step closer to truly changing the way we care for birthing individuals and achieving our goal of ensuring that everyone in our community has a healthy pregnancy and early transition to parenthood.
