More than 10,000 babies are born in the United States every day, but how they’re born isn’t always predictable. Cedars-Sinai investigators developed a machine-learning model that can predict whether a woman will deliver vaginally or by cesarean section. 

"This can help us set expectations with our patients early on in their labor,” says Melissa Wong, MD, MHDS, assistant professor of Obstetrics and Gynecology and first author of the study in the American Journal of Perinatology. "We can prevent a C-section from being performed if a vaginal delivery is likely, or we can prevent a patient from laboring for a long time if the model predicts that a C-section is probable.” 

Significant disparities in outcomes based on race and ethnicity were part of the reason Dr. Wong wanted to develop this tool and hopefully curb unnecessary C-sections, which account for about a third of births in the U.S.—with rates even higher among people of color. While a C-section can be required for a pregnant patient or their baby in many circumstances, it is a serious surgery that carries risks for both.

These disparities posed a challenge in creating an algorithm that would make accurate predictions without building in inherent biases that cause outcome gaps.

Part of the solution was building into the prediction tool the ability to understand how the AI came to its conclusions. For example, if a patient’s likelihood of a vaginal delivery drops from 80% to 60%, clinicians can click on the prediction to see that it’s because the patient’s membranes have been ruptured for a long time and she has developed a fever. 

"You need to be able to ask questions of the model, just as you would a colleague,” Dr. Wong says. "The last thing anybody wants is to feel like we’re strictly making decisions from gut instinct, even the instinct of an algorithm.”