A team of researchers at Seoul National University Bundang Hospital (SNUBH) has developed an artificial intelligence (AI) model to assist in determining the appropriate time for extubating premature infants who are on ventilators after intubation.

The research team, which included Professors Jung Young-hwa and Choi Chang-won from the Department of Pediatrics at SNUBH, as well as Yoo Soo-young and Song Won-geun from the Office of eHealth Research and Businesses at SHUBH, developed a model to assess the success rate of extubation in premature infants on ventilators following intubation. 

Their findings were published in the latest issue of the prestigious journal, International Journal of Medical Informatics.

Premature infants frequently arrive with underdeveloped respiratory systems, increasing the likelihood of breathing challenges and apnea. The level of risk tends to rise with the degree of prematurity, making these babies more susceptible to respiratory complications. In such instances, the child will undergo intubation and ventilation until they can autonomously achieve normal breathing.

It is advisable to use the ventilator for the shortest possible duration and then proceed with extubation. Prolonged use of a ventilator increases the likelihood of abnormal lung injury. Prolonged intubation and ventilator use are known to elevate the risk of bronchopulmonary dysplasia in premature babies and delay their neurological development.

Conversely, premature extubation can result in hypoxia and hypercapnia, negatively impacting the brain and increasing the risk of bronchopulmonary dysplasia. This highlights the importance of timing extubation neither too early nor too late.

Currently, there are no agreed-upon guidelines; instead, the timing of extubation relies on the judgment of the attending physician. On average, there is a success rate of 60-73 percent in maintaining extubation in preterm infants weighing less than 1,000 grams.

The researchers analyzed data from 678 premature babies born at SNUBH between 2003 and 2019 who were intubated and placed on invasive ventilation when they were under 32 weeks of age. They developed a machine learning algorithm to predict the success of extubation by analyzing vital signs such as heartbeat and breathing.

The AI model, named "NExt-Predictor," has demonstrated high predictive capabilities, achieving an area under the curve (AUC) of 0.805 and a precision of 0.917, according to the hospital. This predictive performance remained consistent when analyzed using the US clinical database MIMIC-III. Furthermore, it offers the advantage of not requiring additional equipment, as it relies solely on basic vital signs.

"It is crucial for premature infants on ventilators to be weaned from the ventilator at the optimal time, ensuring it is neither too early nor too late," said Professor Jung Young-hwa. "However, establishing precise criteria for this has proven challenging. That's why we decided to develop an AI system that can predict the likelihood of a successful extubation, a valuable tool in the medical field."

The study is titled, "Development and validation of a prediction model for evaluating extubation readiness in preterm infants."