Researchers at Vanderbilt University Medical Center have developed a way to identify infants at risk for severe neonatal opioid withdrawal syndrome, or severe NOWS, defined as administration of oral morphine.
As reported in Pediatrics, Thomas Reese, PharmD, PhD, assistant professor of Biomedical Informatics, Stephen Patrick, MD, MPH, MS, formerly at VUMC and now chair and O. Wayne Rollins Distinguished Professor of Health Policy and Management at Emory University, and colleagues created and validated a prediction model using electronic health record data from 33,991 births.
For infants with chronic opioid exposure, current guidelines recommend up to seven days of hospital observation. But many exposed infants never develop withdrawal requiring treatment, while others with seemingly less exposure may develop severe symptoms.
Here the team adapts their previously reported predictive model, which had been based on data from Tennessee Medicaid. They used logistic regression to identify seven key predictors available at birth in the health record: maternal opioid use disorder diagnosis and cigarette smoking, Apgar score — a quick, routine assessment of infant health done at birth — and various maternal prescriptions.
In retrospective data, the model demonstrated superior performance compared to standard NOWS screening criteria. A method called the area under the receiver operating characteristic curve, or AUC, gauges a predictive model’s overall clinical usefulness and can be interpreted as measuring accuracy in ranking patients according to risk; with 90% or greater considered outstanding, the team’s model had an AUC of 96%.
The authors are gathering prospective data on model performance preparatory to testing the impact of a decision support tool in a VUMC randomized trial.
Others on the study include Andrew Wiese, PhD, MPH, Ashley Leech, PhD, Henry Domenico, MS, Elizabeth McNeer, MS, Sharon Davis, PhD, Michael Matheny, MD, MS, and Adam Wright, PhD.
The study was funded in part by the National Institute on Drug Abuse (grant P50DA046351) and the Agency for Healthcare Research and Quality.
