Assistant Professor Children's Mercy Hospitals and Clinics Children's Mercy Hospitals and Clinics Kansas City, Missouri, United States
Background: Bronchopulmonary dysplasia (BPD) is the most common respiratory morbidity after preterm birth but requires diagnosis at 36 weeks postmenstrual age (PMA). Dexamethasone is often used to treat infants at high-risk of BPD. The ability for earlier prediction of BPD, based on steroid response, could be useful as a surrogate marker for new therapies.
Objective: To construct a model that predicts severe BPD or death at 36 weeks PMA based on clinical response to dexamethasone.
Design/Methods: Retrospective chart review of preterm infants treated with dexamethasone between 2010-2020 at a Level IV NICU with data collected on demographics, age of steroid initiation, mode and level of ventilatory support during treatment, and pCO2 on day 1, 3, and 7 of steroid use. Highest mode of ventilation was assessed as either high frequency oscillatory ventilation, conventional invasive ventilation, or any form of non-invasive ventilatory support; while support level was represented by respiratory severity score (RSS = MAP*FiO2). BPD outcomes were defined according to the 2017 BPD Collaborative definition. The composite of mild, moderate, or severe (type 1) BPD was used as referent group to assess odds against the composite of severe (type 2) BPD or death.
A regularized logistic model was fitted using the following variables: gestational age, sex, age of steroid initiation, baseline (Day 1) and percent change from baseline (Day 7 vs Day 1) in RSS and pCO2, and ventilator mode change from baseline. The resulting predicted probabilities were divided into quartiles to obtain a discrete risk level (level 1 to 4).
Results: 94 infants were treated with dexamethasone prior to 36-week BPD assessment. A 10,000-iteration bootstrap was performed, and a risk score was predicted for infants not included in the resampled data at each iteration. The proportion of those with severe (type 2) BPD or death at each risk level were evaluated (Figure). For comparison, predictions were also made with a baseline model using only demographic data. Increasing risk category was well aligned with rising outcome incidence, increasing from ~20% of infants at level 1 to just over 55% of infants at level 4. Conclusion(s): The addition of changes in ventilatory parameters with dexamethasone improved BPD prediction compared to baseline demographics alone. Incorporating drug response phenotype into a BPD model may enable more rapid development of future therapeutics.
Acknowledgments: Gary Krueger and Warren Teachout for technical assistance with data collection.
Figure: Proportion of severe (type 2) BPD or death by Risk Score category.
Baseline model includes demographic factors only, Model includes factors quantifying ventilatory change from baseline with dexamethasone treatment and demographics. Boxplots represent proportion of infants with BPD severe (type 2) or death labeled with each risk level by the respective models over 10,000 bootstrap iterations. Open circles represent proportion estimates outside of 1.5 x interquartile range for each model.
Authors/Institutions: Christopher R. Nitkin, Children's Mercy Hospitals and Clinics, Kansas City, Missouri, United States; Keith Feldman, Children's Mercy Hospitals and Clinics, Kansas City, Missouri, United States; Alain Cuna, Children's Mercy Hospitals and Clinics, Kansas City, Missouri, United States; Alexandra Oschman, Children's Mercy Hospitals and Clinics, Kansas City, Missouri, United States; William E. Truog, Children's Mercy Hospitals and Clinics, Kansas City, Missouri, United States; Michael Norberg, Children's Mercy Hospitals and Clinics, Kansas City, Missouri, United States; Jane B. Taylor, University of Pittsburgh Medical Center, Pittsburgh, Pennsylvania, United States; Tamorah Lewis, Children's Mercy Hospitals and Clinics, Kansas City, Missouri, United States