532 - Early Heart Rate Pattern Predicting Cerebral Palsy in Extremely Preterm Infants

Publication Number: 532.4820

Karen D. Fairchild, University of Virginia School of Medicine, Earlysville, VA, United States; Douglas E. Lake, University of Virginia, Charlottesville, VA, United States; Rathinaswamy Govindan, Children's National Hospital, Washington, DC, United States; Lisa C. Letzkus, UVA School of Medicine, Charlottesville, VA, United States; Adre du Plessis, George Washington University School of Medicine and Health Sciences, Bethesda, MD, United States

Background: Identification of preterm infants at high risk for an eventual diagnosis of cerebral palsy (CP) can lead to early interventions during critical periods of brain development to optimize outcomes. Abnormal heart rate variability (HRV) can reflect brain injury and dysfunction, and is typically assessed using inter-heartbeat intervals from ECG waveform data. Using a simpler approach, we found that low variability of HR collected at 0.5 Hz (one value every 2 seconds) measured early in the NICU stay predicted CP (PMID 37891365). We sought to validate this finding on extremely preterm infants and include another NICU for external validation.

Objective: For extremely preterm infants at 2 NICUs, analyze a simple, early HRV metric for predicting CP

Design/Methods: We included infants < 29 weeks gestation at two level IV NICUs (University of Virginia, UVA, and Children’s National Medical Center, CNMC) who survived >14 days and had archived HR data. Medical records were searched for diagnosis of CP and for death after day 14. HR standard deviation (SD) was calculated on 10-minute data segments (300 HR values). Median HR SD was calculated for week 1 and week 2 after birth. “Early HR SD” was assigned to each infant as the lowest of the two weekly values. Using a HR SD threshold from prior work we compared sensitivity, positive predictive value (PPV) and F1 score (harmonic mean of sensitivity and PPV) for CP diagnosis for infants at each NICU. Since birthweight is known to be inversely correlated with CP risk, we calculated cross-validated area under the ROC curve (cvAUC) for birthweight alone, and then using multivariate logistic regression, with early HR SD.

Results: The combined cohort included 467 infants (mean BW 880 grams), of whom 30 (6%) died after day 14 and 73 (16%) were diagnosed with CP (Table 1). In total, over 750,000 10-minute HR records were analyzed. Using an early HR SD threshold of 2.75 beats per minute, sensitivity for CP was 19% and 64%, PPV was 52% and 37%, and F1 was 0.28 and 0.47 at UVA and CNMC, respectively. Figure 1 shows birthweight (x axis) and early HR SD (y axis) for infants with and without CP, and for those that died after day 14, at the two centers. A logistic regression model for CP prediction with birth weight alone had cross-validated AUC of 0.592 on the combined data set. Adding early HR SD to the model significantly increased the performance to 0.666 with a coefficient p-value of 0.002.

Conclusion(s): For extremely preterm infants at two NICUs, a simple measure of low variability of heart rate in the first 2 weeks after birth predicts CP.

Table 1
Demographics and outcomes of cohorts

Figure 1: Early HR SD in relation to birthweight and outcomes.
In the first two weeks after birth, HR standard deviation was analyzed for all infants <29 weeks gestation at two NICUs that survived beyond 14 days and had archived 0.5 Hz HR data available. Data are shown for infants at UVA (left) and CNMC (right). Birthweight for each infant is shown on the x axis and early HR SD on the y axis for infants later diagnosed with CP (large red dots), and those that survived without CP (small blue dots). Infants that died after day 14 are depicted (light gray x’s) but were not included in performance metrics.