Director, pharmacometrics Critical Path Institute, United States
Disclosure(s):
Yi Zhang, PhD: No financial relationships to disclose
Objectives: Bronchopulmonary dysplasia (BPD) is a chronic inflammatory lung disease that affects thousands of neonates and infants every year. The pathophysiology and severity are characterized by the need for supplemental oxygenation or ventilatory support at 36 or 40 weeks post-menstrual age (PMA). In response to a recent grant award from FDA, the International Neonatal Consortium (INC) has an opportunity to collate and evaluate real world data (RWD) sources to assess their value in informing various aspects of neonatal drug development [1]. As part of this effort, our objective is to develop a time-to-event (TTE) model for the off-respiratory support events, to improve understanding of the disease course and predictors of the probability of off-respiratory support events.
Methods: Subject-level data based on electronic health records (EHR) of 1244 babies with BPD were collected. Following exploratory Kaplan-Meier analysis, the model was trained and validated using 10-fold cross validation. The baseline is as the date of birth. Covariate model building is based on scientific plausibility and full-covariate approach by considering both p-values and +-15% intercept value effect size range for statistical significance and clinical significance, respectively. Candidate covariates including sex, birth weight, birth height, gestational age (GA), respiratory support types, and symptoms such as wheezing, tachypnea, patent ductus arteriosus (PDA), among others. Both proportional and non-proportional parametric hazard models were considered for the event and (right) censoring. Model comparison is based on AIC and goodness-of-fit metrics such visual check, Brier score, and c-index.
Results: Both Kaplan-Meier analysis and scientific plausibility suggest non-proportional hazard. Results from parametric survival models suggest that the generalized Gamma survival distribution predicts the out of sample data well, on par with odds-based spline models, though the latter requires more tuning on the location of knots. The final covariate model shows that sepsis and atelectasis lower the off-respiratory-support probability, while the observation of wheezing is related to increased probability of becoming off-respiratory support. GA is also a significant predictor: patients with higher GA tend to be off support sooner. FiO2 as a respiratory measure is mostly of room air value thus not an effective predictor by itself.
Conclusions: To our knowledge this is the first TTE analysis to study off-respiratory support for BPD, and we face many challenges in utilizing RWD, including data curation, accurate definition of events, and covariate missingness. Understanding the predictors in the context of these challenges can help us further examine possible risk factors and prepare for future EHR-based analysis.
Citations: [1] Barrett JS, Cala Pane M, Knab T, et al. Landscape analysis for a neonatal disease progression model of bronchopulmonary dysplasia: Leveraging clinical trial experience and real-world data. Front Pharmacol. 2022;13:988974. Published 2022 Oct 12. doi:10.3389/fphar.2022.988974
