79 - Impact of Race in the HeRO Randomized Controlled Trial

Poster Number: 79
Publication Number: 79.437

Brynne Sullivan, University of Virginia School of Medicine, KESWICK, VA, United States; Camille Hochheimer, University of Colorado School of Medicine, Aurora, CO, United States; Pavel Chernyavskiy, University of Virginia School of Medicine, Charlottesville, VA, United States; William E. King, MPSC, Charlottesville, VA, United States; Karen D. Fairchild, University of Virginia School of Medicine, Charlottesville, VA, United States

Background: A multi-center RCT of 3,003 VLBW infants showed that displaying a heart rate characteristics (HRC) index (HeRO score) predicting late-onset sepsis reduced mortality. The HRC algorithm informs clinicians of the risk of sepsis but the RCT protocol did not mandate action. Differences in its impact might exist due to variability in how clinicians use it to guide decisions and interventions.

Objective: To determine whether HRC was equally accurate in predicting sepsis among Black and White infants and whether randomization to HRC-display had a different impact on clinical interventions or outcomes for Black vs White infants.

Design/Methods: We performed a secondary analysis of data from infants in the HRC RCT with maternal race identified as Non-Hispanic Black or White. We evaluated the calibration of the HRC index separately for Black and White patients using predictiveness curves. Using Bayesian regression models adjusted for birth weight, we assessed multiple outcomes and interventions for the presence of a statistically significant race x randomization interaction. Interventions included the number of blood cultures from 3 to 120 days of age, days on antibiotics, and days on mechanical ventilation. Outcomes included late-onset sepsis, mortality, and length of stay among survivors. We considered 95% Credible Intervals (CrI) that excluded 0 as statistically significant.

Results: In 2,607 infants analyzed, White infants had higher birthweight and gestational age and more ventilator days, while sepsis and mortality were similar(Table 1). The HRC index was well calibrated for sepsis prediction in Black and White infants(Fig 1). Using modeling, we found no differential effect of randomization by race on sepsis, LOS, or mortality. We found no effect by randomization or the interaction term on ventilator days, but there was an effect by race. We found an effect of race on antibiotic days, given that any antibiotics were prescribed. However, there was no differential randomization by race effect(Fig 2a). There was a significant interaction effect of randomization and race on the number of blood cultures per patient, given at least one was obtained. For White infants, the randomization RR was 1.11 (95% CrI 1.04-1.18) compared to Black infants with RR 1.00 (0.93-1.07) (Fig 2b).Conclusion(s): The HRC index is well and equally calibrated for sepsis prediction in Black and White infants. We found a differential effect of randomization by race on the number of blood cultures per infant. For other outcomes or interventions, we found differences by race or randomization, but no interaction between these two variables.
Table 1. Characteristics of Non-Hispanic Black and White infants enrolled in the HeRO trial.Categorical variables are summarized by frequency and percentages and continuous data are summarized by mean and standard deviation or median and interquartile range. All data are truncated at the death, NICU discharge, or 120 days. *Indicates a difference between Black and White by Wilcoxon rank-sum tests or Chi-squared test with p < 0.05. Values reported for “Blood Cultures” represents the median [Q1-Q3] number of blood cultures obtained between day 3 and 120 (or death or NICU discharge) per patient.
Figure 1. Predictiveness curves in Black and White infants randomized to the HRC Display group.Fig_HeRO_race.jpegThe smooth red line and blue dashed line represent the predicted risk of sepsis based on HRC index for Black and White patients, respectively. The red circles and blue triangles indicate the observed rate of sepsis for Black and White patients, respectively, within each decile of predicted risk. The agreement between the predicted and observed rates of infection indicates the HRC index is well calibrated for both Black and White patients.