NICUs And AI For Babies

How Can AI Fit Into the Field of Neonatology?

Preterm births account for about 10% of live births, and almost all preterm newborns have trouble feeding. The number of newborns in the United States who experience difficulty feeding is increasing, and it currently stands at over 2.8 million.

In neonatology, artificial intelligence is currently being tested for a variety of purposes, such as monitoring vital signs, diagnosing and prognosing neurological disorders, and predicting diseases such as respiratory distress syndrome, bronchopulmonary dysplasia, apnea of prematurity, retinopathy of prematurity, intestinal perforation, and jaundice.

A Review of NICU Data Quality

Think about a typical neonatal intensive care unit (NICU): the tempo of care is high, and practically all clinical decisions are made in real-time. The lack of high-quality data is already making those judgments difficult. Despite the wealth of clinical data stored in EHRs, the unstructured data that NICU teams require is frequently obscured, absent, or filled with errors in patient records.

A NICU nurse, for example, can probably tell if a dosage of medication has been inadvertently increased by zero based on a baby’s weight and can immediately account for the mistake because the human eye can fix many of these mistakes. That AI will act similarly is still a mystery to humans. Already, clinicians’ mistrust and the excessive amount of rework caused by poor data quality in NICUs are out of control.

It’s adding fuel to the fire of burnout among nurses who feel a moral imperative to verify every detail in a patient’s chart, even if they rarely have the opportunity to do so. Assuming the AI systems are indeed developed using pediatric data, NICUs are not in a position to responsibly rely on them for any part of patient care.

There is a dearth of pediatric data in current AI research, which causes researchers to draw incorrect conclusions about pediatric populations from adult datasets, as pointed out in a new framework of guidelines for the responsible use of pediatric data in AI studies. It is often believed that NICUs should immediately begin using AI to improve patient outcomes and decision-making. However, health systems would be better served by exercising prudence and bolstering the quality of their clinical data in NICUs. You get out of life what you put into it, as the adage goes.