New prediction models spot PH risk in babies with lung condition
AI tools identify risk in preemies with bronchopulmonary dysplasia
Written by |
Newly developed prediction models can identify infants with bronchopulmonary dysplasia (BPD) — a chronic lung condition that primarily affects premature newborns who need oxygen therapy — who are at the highest risk of developing pulmonary hypertension (PH), according to a large, multicenter study.
After analyzing data from more than 20,000 infants across as many as 228 hospitals, researchers found that lower birth weight and specific clinical markers could help predict which babies face sharply higher risks of PH and death.
“Accurately predicting the outcome of PH will be the first step in identifying the highest risk infants to improve monitoring, selection for possible preventative therapies, and outcomes,” the researchers wrote in the study, “Predicting pulmonary hypertension in infants with bronchopulmonary dysplasia,” published in the Journal of Perinatology.
BPD affects premature infants, especially those who require long-term oxygen therapy or mechanical ventilation. Because their lungs are underdeveloped, these approaches help them grow so that babies can breathe on their own. But the pressure from the oxygen they receive can cause inflammation and scarring, leading to breathing problems such as rapid breathing and wheezing.
In more severe cases, BPD can progress to PH (high blood pressure in the lung blood vessels). PH can strain the right side of the heart and is linked to serious complications, including heart failure, long hospital stays, developmental delays, increased oxygen needs, and death.
Duke research focuses on at-risk preemies
While doctors know that low birth weight, the need for breathing support, and a reduced fraction of inspired oxygen (the amount of inhaled oxygen) raise the risk of BPD, less is known about which babies with BPD are most likely to develop PH.
To address that gap, a team led by researchers at Duke University Medical Center in North Carolina studied infants born at 22-28 weeks of gestation. (Full term is about 40 weeks). They built and tested prediction models using routine clinical data collected in neonatal intensive care units.
The team analyzed 2,849 infants at 33 weeks postmenstrual age (PMA) — the gestational age plus the time since birth — on mechanical ventilation across 170 hospitals. Overall, 360 babies (13%) developed PH, typically diagnosed at a median of 39 weeks PMA.
Babies who developed PH were born earlier (median 25 weeks vs. 26 weeks gestational age) and weighed less than those without PH (641 grams vs. 731 grams). Black infants were disproportionately affected, accounting for 41% of PH cases compared with 28% among those without PH. More infants in the PH group died (23% vs. 7%).
The researchers fed the model demographic and clinical data, and found six key predictors: birth weight, total days of caffeine therapy, total days of systemic steroid use, early signs of PH, the last fraction of inspired oxygen, and carbon dioxide levels on blood gas testing.
The team then used these features in three distinct machine learning models — which are trained on large amounts of data to make predictions — and the models were able to distinguish infants who developed PH from those who didn’t in 63.5%-72.6% of cases.
The models’ performance remained consistent when tested in a separate group of 252 infants across 78 hospitals.
The researchers also examined infants at 36 weeks PMA. At that timepoint, 20,173 premature infants receiving any respiratory support were studied across 228 hospitals, including 2,385 who had also been evaluated at 33 weeks. Among them, 770 babies (4%) developed PH at a median PMA of 40 weeks.
Again, infants who developed PH were more premature (median 25 weeks vs. 26 weeks), had lower birth weights (660 grams vs. 810 grams), and were more likely to be Black (39% vs. 26%). Mortality was also markedly higher among infants with PH (12% vs. 1%).
At 36 weeks PMA, the strongest predictors of PH included birth weight, the presence of an atrial septal defect (a hole in the wall separating the heart’s upper chambers), total days of systemic steroid treatment, respiratory support before 36 weeks PMA, and the most recent blood chloride and carbon dioxide levels.
Here, the three models discriminated between babies who eventually developed PH and those who didn’t with higher accuracy, ranging from 80.3% to 82.6%. When validated in a separate group of 2,749 infants across 163 hospitals, the model continued to perform well.
Accuracy was similar across sex and racial groups and remained stable even after excluding infants diagnosed with PH based solely on inhaled nitric oxide treatment, a PH treatment.
“We have developed and validated predictive models for PH in infants at two timepoints,” the researchers concluded. “These models use readily available clinical variables to assist with bedside clinical decision making.”
