A Replication and Extension of a Prediction Tool Identifying Need for Treatment Among Opioid Exposed Infants

• In Event: Poster Session 02
  In Poster Session: PS 02 Topic - Biological Processes: Psychophysiology

Abstract

The incidences of maternal opioid use and neonatal opioid withdrawal syndrome (NOWS) have increased by nearly 400% over the past decade. Research indicates that between 50 – 80% of infants with prenatal opioid exposure are diagnosed with NOWS (Conradt et al., 2019). Consequently, it is important to predict which infants with prenatal opioid exposure will require pharmacotherapy from those infants that will not require pharmacotherapy. Isemann and colleagues (2017) developed the TiTE and TiTE² to predict the need for pharmacotherapy within the first 36 hours of life. The TiTE measures the average score of three symptoms from the Finnegan (e.g., increased muscle tone, tremors when disturbed, and excoriations) around 36 hours of life. The TiTE² incorporates exposure type into the prediction model to enhance the predictive value. Both prediction tools have high positive predictive values but low sensitivity.

The current study replicated and extended Isemann and colleagues’ (2017) prediction tool. Furthermore, a factor analysis examined which items from the Finnegan loaded together in this specific dataset to improve the sensitivity and generalizability of the prediction tool.

This study utilizes data from a retrospective chart review conducted on all deliveries between 2011 and 2016. All newborns with an ICD 9/10 code for NOWS or with prenatal opioid exposure (as determined by positive UDS or maternal report at delivery) were identified (n = 2638). 380 of these infants had detailed Finnegan scores recorded within the first 48 hours of life. TiTE and TiTE2 scores were calculated based the description in Isemann et al., 2017. A treatment variable was created to indicate whether the infant received pharmacological treatment (1) for their withdrawal or not (0).

Sensitivity, specificity, positive predictive values, and negative predictive values were calculated for the TiTE and TiTE² scores using the cross tabs function in SPSS version 24.0 and are presented in Table 1. This series of experiments successfully replicated Isemann et al., (2017) results and established alternative cutoff values for requiring treatment that provide better balance between all four metrics. However, modifying the polysubstance category to include marijuana and benzodiazepine exposure did not improve sensitivity.

Given that the sensitivity values of the replication experiments were not significantly improved by incorporating additional polysubstance categories or by using alternate cutoff values, we then explored the possibility that the items included in the TiTE prediction model were not suitable for the current dataset. Thus, an exploratory factor analysis was conducted on the 21 individual items from the modified Finnegan.

New prediction models (TEN/TEN2) are described based on a factor analysis of modified Finnegan scores across the first 48 hours of life. Area ROC curve analyses indicated that the TEN2 was the best prediction model of requiring pharmacological treatment compared to the TiTE2 and the TEN (see Figure 1).

The current study replicated and extended the original findings of Isemann et al., (2017) from an urban region to a unique sample from a rural region of Appalachia. Additionally, it reinforced the importance of considering the type of prenatal drug exposure in prediction models.

Authors

• Loni Parrish, East Tennessee State University
  Presenting Author

• Alyson Chroust, East Tennessee State University
  Non-Presenting Author