In an exciting first, researchers at Stanford Medical School and the March of Dimes Prematurity Research Center (PRC) at Stanford have created a Machine Learning (ML) model that predicts prematurity related newborn diseases weeks before they occur - including before a baby is even born.
Using Electronic Medical Records (EMR) of pregnant Californian women and their babies (including information on the mother’s health and lifestyle before pregnancy, her pregnancy medical data, fetal ultrasound data and the baby’s blood work, weight and APGAR score at birth), the model was able to accurately predict a variety of adverse outcomes, including death.
In addition, the machine learning algorithm, which works by ‘reading’ an immense amount of verbal and numerical medical data, was accurate in predicting outcomes for babies before birth, paving the way for a fundamental leap forward in early diagnosis and treatment of prematurity-related diseases.
“What surprised us is that we could make some of these predictions even before birth,” said senior study author Dr. Nima Aghaeepour, an associate professor of anesthesiology, perioperative and pain medicine and of pediatrics at Stanford Medicine. “This tool has given us something that currently does not exist - the ability to see into the future and act today so we can improve the trajectories of premature babies, and in some cases, save their lives.”
The study results, published this week in Science Translational Medicine, must be validated in larger, more diverse patient populations before the model can be commercialized and made available to clinicians in hospitals around the country, a milestone that remains several years away.
Still, the initial results show promise for two reasons. First, the model showed a high degree of accuracy in predicting a number of adverse outcomes, a characteristic that instills confidence in the reliability of the model. Second, the model was validated with a second set of patient data and performed with a high degree of accuracy in predicting future health outcomes for those babies.
“This type of predictive capability has the potential to dramatically alter care for the most at-risk babies,” said Dr. David Stevenson, a neonatologist, study co-author and lead investigator of the Stanford PRC. “It also shifts the clinical emphasis away from making sweeping inferences about the prognosis of premature babies based on what week they were born and onto individual babies with their own distinct health profiles and maternal histories, which we’ve shown have a significant effect on their offspring.”
The research highlighted the role maternal experiences, not just health histories and pregnancy profiles, have on babies. Specifically, the new algorithm was able to link specific types of Social Determinants of Health (SDOH) in mothers with certain prematurity complications. “If a mother was homeless, we found that the impact on the baby would be different than the impact of incarceration, whereas under traditional paradigms both of these socioeconomic factors might have similar effects on prematurity,” Aghaeepour said.
Currently, doctors primarily rely on APGAR scores (which measure a baby’s pulse, muscle tone, skin color and more), gestational age, and birthweight to gauge a baby’s health, neither of which are particularly insightful in making predictions about future health outcomes for individual babies.
“This model represents a scientific crystal ball for doctors, allowing them to see which infants may need immediate intervention,” said March of Dimes Chief Scientific Officer Dr. Emre Seli. “Usually, scientists find ways to improve on something that we can already do, but in this case, Stanford Medicine researchers have made enormous headway toward finding an answer to a problem that no one had ever solved before.”
“We are encouraged by this breakthrough and hopeful about the impact it can make for at-risk mothers and babies in our lifetime."
To arrive at the ML predictions, researchers grouped EMR of mothers presenting at Stanford Health Care with the EMR of their babies born at Stanford Medicine Children’s Health, covering 22,104 live births between 2014 and 2018. (The validation that occurred in the second set of patients brought the total number of mother-baby pairs to 32,354).
The predictive model uses an algorithm called a long short-term memory neural network to make predictions about 24 health outcomes for babies up to eight weeks old. This neural network operates similarly to how you would read a book – you don’t necessarily recall evert word; instead, you recall concepts, themes and key details, adding to the story as you continue reading.
Using data from mothers and babies just born, the model accurately identified babies who would later develop certain conditions, including bronchopulmonary dysplasia, a type of lung disease, retinopathy of prematurity, which can result in blindness and necrotizing enterocolitis, a serious gastrointestinal condition.
The model could also predict some outcomes before birth, including for morbidity. It had moderately strong predictions for a dozen other outcomes, and weaker predictions for conditions such as yeast infections or meconium aspiration syndrome, which is when a baby inhales meconium during birth.
Dr. David Stevenson is the lead investigator at the March of Dimes Prematurity Research Center (PRC) at Stanford University.
Stanford is one of five March of Dimes Prematurity Research Centers conducting translational research into the causes of preterm birth.
The centers - at Stanford (2011), the University of California San Francisco (2022), Imperial College London (2018), the University of Pennsylvania (2014) and the Ohio Collaborative (2013) - are working independently and together to conduct groundbreaking research that will lead to the development of diagnostic and therapeutic tools to predict and prevent preterm birth.
The heart of our research program, our Prematurity Research Centers conduct inquiries that span the field of prematurity and maternal-fetal health, including research on the molecular and genetic mechanisms leading to preterm birth, the maternal microbiome and immune responses, maternal-placental contributions to preterm birth, environmental factors and social determinants, diagnostic biomarkers of preterm birth and maternal co-morbidities.
As the largest - and most interdisciplinary - scientific undertaking related to solving preterm birth, March of Dimes Prematurity Research Centers are at the forefront of breakthroughs that will fundamentally change pregnancy and birth in our lifetime.