An artificial intelligence-based tool can predict the medical trajectories of individual premature newborns from blood samples collected soon after they are born, a Stanford Medicine-led study has shown.
The research, which published Jan. 21 in Science Translational Medicine, provides a new understanding of the complexity of premature birth, not as a single problem defined by early arrival but as several distinct conditions. The study is a step toward predicting and preventing complications of prematurity using treatments tailored to each patient, the research team said.
“It’s very common to see patients who struggle with one prematurity complication but not all of them,” said co-senior author Nima Aghaeepour, PhD, the Anesthesiology, Perioperative and Pain Medicine Professor II as well as a professor of pediatrics and of biomedical data science.
“We can’t just throw everybody born early into one ‘premature’ category,” he added. “Using biological measurements collected at birth, we were able to come up with a new definition for prematurity that is based on the actual outcomes of these babies.”
The study also offers a starting point for understanding the biology of why different prematurity complications develop, as well as opportunities to interrupt specific biological pathways before they derail enough to produce a full-blown prematurity complication.
Mining data from more than 13,000 California preemies, the AI algorithm found patterns in the blood samples that correlated with the babies’ health later in infancy. The study included babies who developed one or more prematurity complication in the weeks after birth as well as babies born equally early who remained complication-free.
“The goal is to have a whole new taxonomy of prematurity, so you can see where a child is headed and understand what is causing differences in their health trajectories,” said study co-author David Stevenson, MD, the Harold K. Faber Professor in Pediatrics. “That will allow us to intervene, prevent and treat complications.”
The study’s co-lead authors are research scientist Alan Chang, PhD, and Jonathan Reiss, MD, instructor in pediatrics. The co-senior authors include Gary Shaw, DrPH, the Rosemarie Hess Professor and professor of pediatrics, and Karl Sylvester, MD, professor of pediatric surgery.
Complications difficult to predict
Babies born more than three weeks early are considered premature. In the days and weeks after birth, some preemies develop complications that can harm the brain, eyes, lungs or digestive system. In general, earlier-born and smaller preemies face more complications, but individuals born at the same stage and weight can have very different medical paths.
At present, doctors struggle to predict who will do well or poorly.
Every premature baby we save creates an entire lifetime of healthy years.”
To untangle this question, the research team used data collected during routine newborn screening. All babies, including preemies, have blood samples collected at birth on a small card to test for molecules that signal rare, severe metabolic diseases. California also offers these samples for research.
The scientists analyzed blood-spot data from 13,536 premature babies born in California between 2005 and 2010 who arrived more than 10 weeks before their due dates, as this group has the most complications. The researchers also had diagnostic codes from each baby’s medical records: The data included infants with one or more of the four major complications of prematurity — necrotizing enterocolitis, an intestinal complication; retinopathy of prematurity, an eye disease; bronchopulmonary dysplasia, a lung problem; and intraventricular hemorrhage, a form of brain bleeding. The group also included preemies born more than 10 weeks early who developed no complications. The AI algorithm was validated with data from an additional group of 3,299 preterm babies born in Ontario, Canada.
The blood spot data includes measurements of molecules that signal metabolic diseases, such as levels of different amino acids (protein building blocks) and of molecules generated when fats are broken down for energy.
The AI algorithm found patterns in these molecules that correlated with an infant later developing one or more of the four major prematurity complications.
From this, the scientists built a set of six blood measurements that form the basis of a metabolic health index showing which preemies are and are not at risk for complications. The team added basic clinical factors to the index, including the stage of pregnancy at birth, birth weight, infant sex and Apgar scores. With the blood spot data and clinical factors, the index could predict development of each of the four major prematurity complications with greater than 85% accuracy.
The researchers are expanding the AI model by incorporating additional data, such as information from the mother’s pregnancy, data from the baby’s electronic health record and other biological measurements.
Predicting prematurity complications could have many practical advantages, such as guiding which preemies need to be transported to hospitals with high-level neonatal intensive care units and helping physicians give parents better information about their baby’s prognosis.
“It’s a complete change in the way we think about prematurity,” Stevenson said.
Added Aghaeepour, “The opportunity to create value for society is truly outstanding. Every premature baby we save creates an entire lifetime of healthy years.”
Besides the ability of the findings to predict prematurity complications, the research team is excited about expanding scientists’ understanding of how prematurity complications occur.
Although certain clinical factors are generally linked to more complications, “We never knew what was under the hood, so to speak,” Stevenson said. “Now we’re literally looking at the biological machinery and how it’s working.”
Researchers from the University of California, San Diego; the University of Ottawa; the Ottawa Hospital Research Institute; and the Bruyère Health Research Institute contributed to the study.
Funding was provided by the National Institutes of Health (grants R35GM138353, R01HL13984404, 949 R01AG058417 and R61NS114926), the Burroughs Wellcome Fund, the Charles and Mary Robertson Foundation, the Bill and Melinda Gates Foundation, the March of Dimes, the Canadian Institute for Evaluative Sciences, the Ottawa Hospital Research Institute, the Ontario Ministry of Health, and the Ontario Ministry of Long-Term Care. At Stanford University, the Departments of Pediatrics; of Surgery; and of Anesthesiology, Pain and Perioperative Medicine provided funding along with Stanford Medicine’s Metabolic Health Center, its March of Dimes Prematurity Research Center, and the Maternal and Child Health Research Institute.
