252 - Predicting Need for NICU Level Intervention for Late Preterm Infants

Publication Number: 252.6410

Neha S. Joshi, Stanford University School of Medicine, Stanford, CA, United States; Jochen Profit, Stanford University School of Medicine, Palo Alto, CA, United States; Yuan Gu, Stanford University, Palo Alto, CA, United States; Henry C. Lee, University of California, San Diego School of Medicine, La Jolla, CA, United States

Background: While some infants born at late preterm gestation (34-36 weeks’ gestational age, GA) can thrive in well newborn care, this subset is at higher risk of needing neonatal intensive care unit (NICU) level intervention compared to term counterparts. Admission criteria vary by institution and are mostly commonly GA and birth weight (BW) based; there are no evidence-based criteria to determine appropriate admission location for late preterm infants.

Objective: To identify predictors of needing NICU level intervention in late preterm infants.

Design/Methods: We utilized a retrospective cohort of 1022 infants born at late preterm gestation between 2019-2021 at a single institution for model building; cohort details are described in previous literature. Candidate predictor variables were available to clinicians within 2 hours of an infant’s birth to determine admission location; these were GA, BW, delivery mode, sex, 1 minute APGAR, 5 minute APGAR, hypoglycemia (≥1 glucose < 45 mg/dL), tachypnea (≥1 respiratory rates > 60 breaths/minute), and hypothermia ≥1 temperature < 36.5C). A univariate analysis compared candidate predictor variables between infants receiving and not receiving NICU level interventions. Any covariates with p≥0.2 in the univariate association were then excluded. We generated receiver operating curves (ROC) for GA and BW to determine optimal cut off points before dichotomization of GA/BW for multivariable models. We then developed multivariable regression models using stepwise Akaike information criterion, and calculated the sensitivity, specificity, positive likelihood ratio, and negative likelihood ratio.

Results: Following univariate analysis with candidate predictor variables, delivery mode and 1 minute APGAR were removed given p≥0.2. ROCs with the highest area under curve (AUC) included (A) 35 weeks’ GA and 2400 grams BW with AUC 0.770, and (B) 36 weeks; GA and 2400 grams BW with AUC 0.760. Multivariable regression analyses with GA and BW dichotomized for (A) and (B) are shown in Table 1. Diagnostic testing for proposed admission criteria thresholds is shown in Table 2.

Conclusion(s): There is limited data available shortly after birth to help determine admission location for late preterm infants. While GA, BW, 5 minute APGAR, tachypnea, and hypoglycemia may help predict need for NICU level intervention, the overall sensitivity and specificity of using these to model need for NICU level intervention is low. Further work can expand to include a diverse cohort across multiple institutions to mitigate bias to this model from its single institution data.

Table 1
LPIPredict_Table1PAS.pdfTable 1. Multivariable regression model with gestational age and birth weight dichotomized according to highest AUC from ROCs. Table 1A shows dichotomization at threshold of 35 weeks’ gestational age and 2400 grams birth weight and table 1B shows dichotomization at threshold of 36 weeks’ gestational age and 2400 grams birth weight.

Table 2
LPIPredict_Table2PAS.pdfTable 2. Diagnostic testing of proposed admission criteria thresholds.