Can AI help doctors predict and prevent preterm birth?

A machine learning approach promises to help solve the problem

Feb 06 2020 | By Bernadette Young

Ansaf Salleb-Aouissi.

Almost 400,000 babies were born prematurely—before 37 weeks gestation—in 2018 in the United States. One of the leading causes of newborn deaths and long-term disabilities, preterm birth (PTB) is considered a public health problem with deep emotional and challenging financial consequences to families and society. If doctors were able to use data and artificial intelligence (AI) to predict which pregnant women might be at risk, many of these premature births might be avoided.

“Premature birth prediction has been an exceedingly challenging problem,” said Ansaf Salleb-Aouissi, a senior lecturer in discipline from the computer science department. “But we are now at a point where we can use machine learning to develop a dynamic risk prediction system for pregnant women. Creating a system that can process large models of data with AI algorithms we develop would be a great benefit to supplement physicians’ ‘real-life’ expertise.”

A recent $1,000,000 grant from the National Institutes of Health National Library of Medicine (NIH-NICHD) will support Salleb-Aouissi’s research to integrate large data-sets with machine learning to help identify pregnant women at risk. Working with a team of researchers from four disciplines: machine learning, genetics, sequential decision-making, and obstetrics and gynecology, Salleb-Aouissi will develop predictive models for PTB that could be used to assist physicians in identifying women at risk who require intervention and those at lower risk who need no intervention.

First-time mothers, because they have no prior pregnancy history, are a particularly challenging population to determine PTB risk. These mothers, known medically as nulliparous women, represent 40% of pregnant women and are simply not treated. A reliable PTB prediction system could substantially reduce their risk.

PTB poses a number of analytic challenges. Only about 10-12% of mothers give birth prematurely, providing a limited number of cases to study. There are several different categories of PTB by gestational age at birth: extreme (<28 weeks), severe (28-31 weeks), moderate (32-33 weeks), and late (34-36 weeks). The team needs to consider spontaneous PTB and indicated PTB separately.

In addition, because pregnancy is a dynamic and fast-changing process and different gestational ages at birth are believed to be driven by different etiologies, longitudinal prediction models are needed. And the attributes of each pregnancy involves time-dependent information that only gradually becomes available for each patient.

The study addresses three important gaps in the current literature:

  • To study nulliparous women and their risk for PTB;
  • To combine genetic factors that have been lacking in most studies and never combined with other risk factors to determine risk; and
  • To use longitudinal data and models to optimize scheduling of patient visits, testing, and treatment.

The interdisciplinary team includes associate professor Itsik Pe’er, along with professor Anita Raja from Hunter College, and Ronald Wapner and his team from the department of Obstetrics and Gynecology, Maternal-Fetal Medicine at Columbia University Irving Medical Center (CUIMC). They will be working with the largest cohort of nulliparous patients to date, using the NIH-NICHD’s recently released nuMoM2b (Nulliparous Pregnancy Outcomes Study Monitoring Mothers-to-be) dataset. The nuMoM2b study focused on a meticulous, prospective cohort study of a racially, ethnically, geographically diverse population of 10,038 nulliparous women with singleton gestations. It also includes genetic information that was never used before.

“The next step is to massage the nuMoM2b data—clean it and bring it into a form amenable to machine learning,” said Salleb-Aouissi, who first became interested in applying machine learning techniques to healthcare when her second child developed colic. This inspired her to apply for and win a RISE grant in 2011 that used machine learning techniques to better understand the causes of colic. That project started a collaboration with researchers from CUIMC that led to an initial study on PTB through a NSF EAGER award. Salleb-Aouissi worked with a dataset of 3,000 cases and was able to correctly predict preterm births about 50% to 60% percent of the time.

“It is our aim to conduct machine learning research that will benefit maternal-fetal medicine. Being able to accurately predict women at risk of PTB could change the way pregnant women are followed and treated,” Salleb-Aouissi added. “Our research would also be applicable to address other adverse pregnancy outcomes such as intrauterine growth restriction, pre-eclampsia, and ultimately other women's health problems.”

Stay up-to-date with the Columbia Engineering newsletter

* indicates required