419 - Clinical and Demographic Predictors of the Need for Pharmacotherapy in Neonatal Abstinence Syndrome (NAS)

Poster Number: 419
Publication Number: 419.136

Shawana Bibi, Cleveland Clinic Children's / Tufts Graduate School of Biomedical Sciences Clinical and Translational Science Program, Solon, OH, United States; Rachana Singh, Tufts Childrens Hospital, Boston, MA, United States; Janis Breeze, Tufts Medical Center, Boston, MA, United States; Jason Nelson, Tufts Medical Center, Boston, MA, United States; Walter K. Kraft, Sidney Kimmel Medical College at Thomas Jefferson University, Philadelphia, PA, United States; Jonathan M. Davis, Tufts Childrens Hospital, Boston, MA, United States

Background: Timely and accurate prediction of the need for pharmacotherapy (PT) for NAS remains elusive. Significant variation in clinical expression of NAS in terms of onset, severity, need for treatment, and duration of physiologic instability has resulted in prolonged birth hospitalization, with a significant increase in the healthcare costs.

Objective: To develop and validate a clinical prediction model of need for PT in NAS using patient specific clinical and demographic data.

Design/Methods: Pooled patient level data was used from 4 separate cohorts: 1) a multisite randomized clinical trial (RCT) and prospective observational cohort, 2) a single center RCT , 3) a prospective observational cohort and 4) a retrospective community birthing hospital cohort. Infants ≥37 weeks gestation born to mothers with history of opioid use disorder (OUD) during pregnancy were included. Primary outcome was need for PT utilizing modified Finnegan scoring criteria (FNASS). Multiple imputation was performed under assumption of missing at random  to account for missing data. A multivariable logistic regression model was built using 12 candidate variables available at the time of admission and employing backward variable selection. Model was internally validated by boot strapping. Leave-one-out validation was performed using data from 3 cohorts to develop the model and external validation on the 4th cohort. This allowed us to examine stability of validation while also performing external validation. Model performance was evaluated using discrimination (area under the receiver operating curve, AUC) and calibration.

Results: 730 infants were included, of which 442 needed PT (Table 1). The primary outcome frequency (i.e. need for PT) across the 4 study cohorts ranged from 53 to 65%. The model AUC was 0.69 (95 % CI: 0.64 - 0.72; optimism corrected 0.66 via boot strapping) yielding 7 predictors of the need for PT including: gestational age, any breast milk, type of maternal opioid for treatment of OUD, exposure to heroin, cocaine, benzodiazepines, and/or antipsychotic medications (Table 2).Conclusion(s): Our results are consistent with existing clinical models in a geographically diverse cohort. Currently available predictive models of NAS are not sufficient to accurately determine opioid exposed infants at high vs low risk of requiring pharmacotherapy. There is a significant need for identification of novel physiologic and genomic biomarkers to enhance the predictive performance of models for need of PT in NAS.
Table 1: maternal and infant characteristics and co-exposures by need for NAS pharmacotherapy
Table 3: Multivariable logistic regression model