ML Tool Identifies Biomarkers of Neonatal Opioid Withdrawal Syndrome

A new study published in JAMA Network Open last week described how a machine-learning (ML) tool accurately identified a set of biomarkers for neonatal opioid withdrawal syndrome (NOWS) using newborn acoustic cry analysis.

According to a clinical report from the American Academy of Pediatrics, NOWS is a withdrawal syndrome that occurs in newborns shortly after birth as a result of opioid use during pregnancy. The prevalence of the condition is growing in the wake of the opioid crisis.

The National Center on Substance Abuse and Child Welfare states that NOWS, also known as neonatal abstinence syndrome, an umbrella term describing prenatal exposure to various substances, including opioids, may include symptoms such as respiratory problems, difficulty feeding, seizures, severe irritability, and high-pitched crying.

According to the study, crying is a distinctive component in scoring tools used to assess infants for NOWS, but it is inadequately measured because many characteristics of crying cannot be detected by human perception. Infant cry characteristics reflect opioid receptor expression through the involvement of the brain stem, which can affect the vocal tract, respiratory system, and gut, the study authors stated.

Variations in the acoustics —defined as the physical properties of sound — of infant cries have also been associated with the gene expression related to a stress response. These genetic pathways determine the regulatory behavior associated with crying, such as the consolability of the infant and the pitch of its cries, which cannot always be detected accurately by human hearing.

To objectively measure these aspects of newborn crying as they relate to NOWS, the researchers sought to develop an ML model that could use newborn acoustic cry analysis to capture the pathophysiological features of withdrawal accompanying the condition.

To develop their tool, the research team recruited 177 full-term neonates from Women & Infants Hospital of Rhode Island between Aug. 8, 2016, and March 18, 2020, who had either been exposed or not exposed to opioids. Cry recordings were processed for 118 neonates, with 65 included in the final analyses.

Neonates exposed to opioids were monitored for signs of NOWS using the Finnegan Neonatal Abstinence Scoring Tool (FNAST), the gold standard for NOWS assessment. The assessment was administered every three hours as part of a five-day observation period, during which audio was recorded continuously to capture crying. The crying of non-exposed neonates was recorded during routine handling before hospital discharge.

From this, 775 hours of audio were collected, which were trimmed into 2.5 hours of useable cries. These were then acoustically analyzed. ML methods were used to identify relevant acoustic parameters and predict pharmacological treatment for NOWS.

The ML-based tool achieved high performance in predicting receipt of pharmacological treatment for NOWS, with an area under the curve of 0.90, an accuracy of 0.85, a sensitivity of 0.89, and a specificity of 0.83.

These findings indicate that newborn cry acoustics have the potential to serve as an objective biobehavioral marker of NOWS and that acoustic cry analysis using ML could improve the assessment, diagnosis, and management of NOWS and facilitate standardized care for affected infants, the authors stated.