We developed prediction models for NAS based on a set of 30 demographic and antenatal exposure covariates collected during pregnancy. Data (outpatient prescription, vital, and administrative records), were obtained from enrollees in the Tennessee Medicaid Program from 2009 to 2014. Models were created using logistic regression and backwards selection based on improvement in the Akaike information criterion, and internally validated using bootstrap cross-validation.
A total of 218,020 maternal and infant dyads met inclusion criteria, of whom 3,208 infants were diagnosed with NAS. The general population model included age, hepatitis C virus infection, days of opioid used by type, number of cigarettes used daily and the following medications used in the last 30 day of pregnancy: bupropion, anti-nausea medicines, benzodiazepines, anti-psychotics, and gabapentin. Infant characteristics included birthweight, small for gestational age and infant sex. A high-risk model used a smaller number of predictive variables. Both models discriminated well with a AUC of 0.89 and were well-calibrated for low-risk infants.
We developed two predictive models for NAS based on demographics and antenatal exposure during the last 30 days of pregnancy that were able to risk stratify infants at risk of developing the syndrome.
Copyright © 2020 Elsevier Inc. All rights reserved.