Maternal and baby medical data predict complications in premature babies, Stanford Medicine-led study shows | News Center

“There are computational challenges to using electronic medical records because they are long-term and contain large amounts of data from each patient,” said Aghaeepour. “Long-short-term memory neural networks work in a similar way to people who read books. Add Key Concepts and Advance It While the algorithm may not remember every patient’s entire electronic medical record, it can remember key concepts and advance them to the point of making predictions. can.”

At birth, machine learning models have provided strong predictions that infants will develop various conditions such as bronchopulmonary dysplasia, a type of chronic lung disease. Retinopathy of prematurity, a problem in the retina that causes blindness and blindness. Anemia of prematurity; necrotizing enterocolitis is a serious gastrointestinal complication that is often not diagnosed until the first few weeks of life, complicating interventions by the first few weeks of life and leading to a poor prognosis.

The model also showed strong predictions for multiple outcomes such as mortality and retinopathy of prematurity, which can lead to vision loss and blindness, and 11 other conditions, one week before birth. showed moderately strong predictions.

“I was amazed at how much predictive power we had before our babies were born, and right after they were born,” Agaeepour said. I thought that collecting data from babies would improve accuracy in the first few days of life.”

Some complications were not reliably predicted by the model, such as which infants developed candidiasis and yeast infections. Polycythemia, high concentration of red blood cells in the blood. In meconium aspiration syndrome, an infant inhales meconium, a sticky substance that is expelled from the fetal intestine at birth.

The researchers verified that the predictive strength remained unchanged over the years (compare 2014-2018 births with 2019-2020 births). They also used an independent group of her 12,258 mother-infant pairs at UC San Francisco to validate some of the findings.

Model predictions at birth provided more accurate information than currently used risk assessment tools such as the Apgar score and the National Institute of Child Health and Human Development risk scores. The researchers note that these scores only consider the condition of the baby at birth and do not incorporate information from the mother’s medical history. But additional studies in more diverse populations are needed before this machine-learning tool is ready to replace existing bedside risk calculators, the researchers said.

mother’s health is important

According to the researchers, the model revealed an unexpected relationship between certain health conditions or social conditions of mothers and the health of their infants.

For example, mothers with anemia (a common pregnancy complication) were more likely to give birth to anemic newborns. The study also found that.

“We need to investigate the linkages that explain these relationships at the biological level, as they may provide clues to how certain conditions arise,” Stevenson said. “Then we will be able to intervene more appropriately to help children.”

The new algorithm was also able to associate certain types of socioeconomic disadvantages in mothers with certain complications of prematurity in babies.

“If the mother was homeless, we found that the health effects on the baby differed from those of confinement, whereas under the traditional paradigm both of these socioeconomic factors had similar effects on prematurity risk.” It may have been thought that it would give a

Predictions from this model could help neonatologists better identify patients who would benefit from existing protocols to prevent complications during childbirth, Stevenson said. For example, newborns who experienced oxygen deprivation at birth can now receive a cooling protocol that lowers their body temperature for several days to prevent brain damage. He said it could help identify

This study needs to be replicated in a larger and more diverse patient population and folded with other Stanford medical studies that characterize pregnancy according to the thousands of biomarkers that change during pregnancy.

Scientists at the University of California, San Francisco contributed to this study.

Funding for the research was provided by the National Institutes of Health (Grants 1R01HL139844, 3P30AG066515, R35GM138353, 1R61NS114926, 1R01AG058417, R01HD105256, P01HD106414, T32GM007618 and T32GM067547), The American Fund of Burrough Wells, The Burrough Wells. Foundation, Alfred E. Mann Foundation, National Science Foundation.