Research: Machine Learning Can Identify Maternal Heart Disease Risk Earlier in Pregnancy.

Research: Machine Learning Can Identify Maternal Heart Disease Risk Earlier in Pregnancy.

Share this

MedStar Health Research Institute’s study suggests a machine learning algorithm could help identify patients at risk of cardiovascular disease earlier in pregnancy, opening a window to intervene.


Our research team assessed a new machine learning algorithm and determined that it can spot birthing individuals’ risk of cardiovascular conditions an average of 56.8 days before typical diagnosis! 

With grant funding from the National Institutes of Health, MedStar Health Research Institute evaluated a tool by Invaryant called Healthy Outcomes for All Pregnancy Experiences – Cardiovascular Risk Assessment Technology (HOPE-CAT). Our research, which has been accepted for publication, shows this technology can identify heart risk in pregnancy early so providers can help sooner.

The U.S. has the highest rate of deaths during pregnancy of any high-income nation in the world. The 2021 rate of 32.9 deaths per 100,000 live births is more than 10 times the rate of other similar nations.  

Machine learning has helped find patients at high risk for other complications and predicted readmission for blood pressure-related complications. In our study, we are among the first to investigate how this technology can be applied to understanding maternal heart risk, which is the leading cause of death among pregnant women.


Preparing the data and training the algorithm.

To study HOPE-CAT, we collaborated with to identify two categories of risk factors for the algorithm to assess:

  • Static: Factors that remain the same, such as race, ethnicity, age, medical history, and family
  • Variable: Things that change more frequently, such as blood pressure, heart rate, and symptoms like headache and shortness of breath

The algorithm used these factors and the anonymous electronic health records (EHR) of 6,069 patients ages 18-40 with more than one pregnancy-related visit who delivered between 2017 and 2020. HOPE-CAT created two risk profiles: Standard risk and high risk. Specific indicators of high risk include:

  • High resting heart rate
  • High systolic blood pressure
  • High respiratory rate
  • Low oxygen saturation
  • Dyspnea (shortness of breath)
  • Orthopnea (difficulty breathing when lying down, but not when standing)

As soon as HOPE-CAT was trained, we cleaned and standardized the EHR data from different records systems into a single virtual server environment. The algorithm examined this data one encounter at a time, from a patient’s first visit through delivery and beyond. 

When HOPE-CAT identified a risk, it generated a profile for that patient and encounter. Linking this profile to the anonymous patient outcomes of cardiovascular conditions allowed us to compare the results. For example, if it detects that a patient had symptoms of kidney disease on day 114 and the EHR shows a diagnosis of kidney disease on day 144, the difference is 28 days. We conducted in-depth manual reviews to cross-check each result. 

Results: Maternal cardiovascular risk identified sooner.

The outcome of our research is clear: HOPE-CAT can effectively identify the risk of cardiovascular disease an average of 56.8 days earlier than the first date of diagnosis. How much earlier depends upon the condition:

Among the 5,238 patients with one or more risk factors, HOPE-CAT identified 1,716 high-risk profiles. Of these, 604 de-duplicated records had cardiovascular outcomes. 

AI could help improve maternal health outcomes.

Early and effective heart risk screening with machine learning technologies could help improve pregnancy outcomes. The earlier we can identify risk, the sooner specialists can help. In places with fewer healthcare resources, earlier identification of risk could mean earlier referrals to specialists.  

HOPE-CAT and technologies like it could help providers deliver more data-driven care, reducing patient morbidity and costs associated with hospitalization while lowering stress on healthcare systems.

More work is needed to ensure this tool is effective in a real-world setting, so that additional research will be done. These exciting results provide foundational evidence that machine learning technologies could unlock solutions to improve birthing patients’ heart health.

Want more information about the MedStar Health Research Institute?

Discover how we’re innovating for tomorrow.

Explore With Us

Stay up to date and subscribe to our blog

Latest blogs