Skip to content

Using Artificial Intelligence to Decrease Obstetric Risk

In This Article

  • Massachusetts General Hospital physician-investigator Mark Clapp, MD, MPH, is developing methods for predicting and avoiding adverse outcomes in high-risk patients
  • Researchers are using artificial intelligence to collect and process historical patient data and generate models capable of stratifying patient risk
  • An unbiased assessment of risk for adverse delivery outcomes ensures that deliveries occur at appropriate facilities able to mitigate risks inherent in the birthing process

Physician-researchers at Massachusetts General Hospital are addressing the underlying causes of high rates of maternal mortality and morbidity. They are applying artificial intelligence (AI) methods for risk stratification in order to optimize labor and delivery outcomes.

"My engineering and public health backgrounds focused on implementing policies to improve quality," says Mark Clapp, MD, MPH, a maternal-fetal medicine specialist in the Department of Obstetrics and Gynecology at Mass General. "As a clinician and researcher, my goal is to identify methods that provide optimal care resulting in the best possible outcomes in obstetrics."

Stratifying Maternal Risk Using AI

The use of AI in medicine has exploded, especially in areas related to obstetrics and gynecology. These applications include the generation of models that allow clinical teams to predict and recognize risk in a given patient—particularly in a labor and delivery setting—in order to mitigate adverse outcomes.

In a recently published study in the American Journal of Obstetrics and Gynecology, Dr. Clapp and colleagues applied a machine learning method called natural language processing (NLP) to analyze large amounts of provider-documented data recorded upon patient admission for delivery. NLP allows a computer to 'understand' the spoken or written word according to a 'language' generated from data. The result is a model comprising a series of linguistic rules specific to the data. Once created, the model can be applied to interpret written documentation and provide predictive probabilities of a particular clinical profile or outcome for a given patient.

In Dr. Clapp's study, the data transformed information from patient charts into statistical probabilities applied to specific words and phrases. Their retrospective analysis demonstrated the utility of this method to accurately predict severe maternal morbidity.

"Our goal with these models is to assign a risk level of maternal morbidity upon admission in order to flag patients that require special care," explains Dr. Clapp.

This assessment is currently undertaken by caregivers already overloaded with both patient and non-patient tasks that can introduce biases into what they see and read. "Our focus is to automate at least a portion of that assessment in order to optimize labor and delivery outcomes."

Recognizing AI-specific Challenges and Capitalizing on Opportunities

NLP generated models are only as good as the data on which they are trained. Because documentation standards and practices can vary widely according to both institution and caregiver, the challenge is how to generalize a model that understands different charting styles.

To address this, Dr. Clapp acknowledges the benefit of being able to use data from throughout the Mass General Brigham network. "This capability allows us to build the models using data from one hospital and validate its predictive accuracy against data from another hospital."

Other NLP-specific challenges include the possibility of worsening underlying health disparities owing to how a model is trained. "If the data used to train the model are heavily skewed toward a particular demographic, disparities can potentially be worsened by applying that model inaccurately to other demographics," Dr. Clapp explains.

Despite these challenges, Dr. Clapp emphasizes the potential of these tools to improve obstetrics care. Specifically, such applications can be invaluable in a highly dynamic environment that requires clinicians to simultaneously balance the risks and benefits of interventions for both mother and child.

"Automating even a portion of that decision-making workload can allow a sharper focus on ensuring successful outcomes," says Dr. Clapp. "In the context of prenatal care, recognition of potential risk in a patient for adverse delivery outcomes can ensure that deliveries occur at the appropriate facility with the resources available to de-risk the birthing process."

Innovation Through Collaboration

Dr. Clapp recognizes the support received from Mass General for this research, as well as the access to colleagues both within and outside of the hospital. Collaborations with members of the Center for Quantitative Health in the Department of Psychiatry, specifically Roy Perlis, MD, MSc, director, and Thomas McCoy Jr., MD, director of research, have been particularly valuable in developing AI applications for use in a clinical setting.

Dr. Clapp has been contacted by local technology companies engaged in either AI or obstetrics research who express an interest in collaborating. "Such partnerships are mutually beneficial, in that we gain colleagues with different backgrounds and technical expertise while providing them with clinical insights and data to drive the field forward."

Although the implementation of this application in a clinical setting will ultimately require approval by the Food and Drug Administration, the biggest hurdle might be acceptance by other clinicians. Dr. Clapp admits that demonstrating evidence of efficacy could be particularly challenging, especially in situations where the model disagrees with the judgment of the clinician.

Nevertheless, he remains confident in the direction the research is moving and its potential impact. "Both Mass General and the Department of Obstetrics and Gynecology have demonstrated a belief in the mission and objectives of this work. That translates to an environment that promotes continually striving to improve patient outcomes, which is truly what drives innovation in this area."

Learn more about the Department of Obstetrics & Gynecology

Refer a patient to the Department of Obstetrics & Gynecology

Related topics


Maternal-fetal medicine specialists at Massachusetts General Hospital found that the Expanded Obstetric Comorbidity Score (EOCS), a risk-stratification tool recently developed for predicting severe maternal morbidity, can also be used to predict which women in labor are at risk of postpartum hemorrhage.


Mark A. Clapp, MD, MPH, a Maternal–Fetal Medicine Program specialist, and colleagues created a computer algorithm to predict severe maternal morbidity simply by reviewing the free-text history and physical notes in the patient's chart at the time of hospital admission for delivery.