654 - Linking Perinatal Acidosis and Bilirubin: A Causal Inference Approach to Neonatal Oxidative Stress

Publication Number: 1631.654

Mia DiCara, Ann & Robert H. Lurie Children's Hospital of Chicago, Long Grove, IL, United States; Kenny Kronforst, Ann & Robert H. Lurie Children's Hospital of Chicago, Chicago, IL, United States; Catherine Berger, Ann & Robert H. Lurie Children's Hospital of Chicago, Chicago, IL, United States; Spencer M. Millen, Northwestern University The Feinberg School of Medicine, Chicago, IL, United States

Background: Infants exposed to perinatal oxidative stress are at increased risk of hypoxic-ischemic encephalopathy (HIE) and resultant complications. Prior models describe bilirubin as a biomarker of HIE clinical outcomes, but few incorporate early markers of metabolic acidosis. Clinical experience suggests a potential causal link between severe early acidosis and bilirubin trends. Building on prior studies that focused on well infants, our study evaluates this association in a cohort exposed to oxidative stress.

Objective: To evaluate the relationship between early base deficit (BD) and peak total bilirubin (TB) using a causal inference framework.

Design/Methods: We identified infants ≥35 weeks' gestational age (GA) admitted to our level III NICU from 2018-2025. Data was obtained from Northwestern's Electronic Data Warehouse. Infants transferred to our level IV NICU were excluded. Infants included needed to have 2+ serum TB measurements and a blood gas within the first three hours of life. Covariates were selected based on clinical relevance related to bilirubin metabolism and included GA, APGAR score, birthweight z-score (Fenton, 2025), hour of life (HOL), hematocrit, base deficit (BD), and therapeutic hypothermia (TH) (Figure 1). Linear regression models using z-scores quantified relationships between first BD peak bilirubin. Three models - unadjusted (Model I), adjusted without TH (Model II), and adjusted including TH (Model III)-- estimated total and direct effects. Analyses were completed in Rstudio (Version 2025.05.0+496).

Results: Of the 8,050 infants born during the study period of 2018-2025, 2,798 met the inclusion criteria. First drawn bilirubin, bilirubin at 24HOL, peak bilirubin, and phototherapy requirement all demonstrated an inverse relationship with BD (Figure 2). This relationship is visible via scatterplot smoothing (Figure 2) and unadjusted linear regression (β = -0.23, p-value < 0.001, Model I, Table 1). To estimate the Total Effect of BD on peak bilirubin, we preserved the indirect pathway between BD and the clinical decision to cool; therefore, Model II does not adjust for TH as a covariate (β = -0.14, p-value < 0.001). Finally, to achieve the Direct Effect of BD on peak bilirubin, TH was included as a covariate (Model III). Indeed, the Direct Effect with the inclusion of TH as a covariate decreased the estimate of BD influence on peak bilirubin (β = -0.08, p-value < 0.001).

Conclusion(s): In infants ≥35 weeks' gestation, initial base deficit may serve as a predictor of peak bilirubin levels. Studies using bilirubin as a marker of HIE severity may consider initial BD as a possible confounder.

Figure 1. Directed Acyclic Graph (DAG) depicting Total Effect of Base Deficit on Peak Bilirubin.
White circles are adjusted in model II. Grey circles are unobserved or unadjusted. Base Deficit is the Exposure in green. Peak Bilirubin is the Outcome in blue. Paths of Total Effect between exposure and outcome are depicted in green arrows. Created with DAGitty (https://www.dagitty.net, Accessed: November, 2025).

Figure 2. Scatterplot and table depicting relationship between first BE and bilirubin in for infants ≥35 week gestational age.
LOESS (locally estimated scatterplot smoothing) lines were drawn only if ≥30 samples and ≥30 patients were available within a 12-hour window.

Table 1. Univariate and multivariate linear regression comparing the exposure of first base deficit (BD, z-score) with the outcome of peak bilirubin (z-score)
Model I is run without adjusting for any covariates. Model II is adjusted for covariates without adjusting for TH, reflecting the Total Effect of BD on peak bilirubin. Model III covariates include TH and represents the Direct Effect. Of note, the N drops for Model II and III because not all patients had two hematocrits available before peak bilirubin.