Methods:
The two-year test cohort consisted of 204 infants born < 30w GA admitted to the NICU at the Medical University of South Carolina in 2015 and 2016. The cut-off of 30w GA was derived from earlier work that eliminated infants who received a GT for non-prematurity-related reasons, such as congenital anomalies.4 Infants who died before discharge or who were admitted after 7 days of age were excluded. We collected retrospective data including demographics, feeding, and other disease variables which are reported elsewhere.4 Respiratory data were collected at three time points in the original cohort: 30 days (d) of age, 32w PMA and 36w PMA. Variables included FiO2 and mode of respiratory support: none, low-flow nasal cannula (NC), heated humidified high-flow NC (HFNC), continuous positive airway pressure (CPAP), nasal intermittent positive pressure ventilation (NIPPV), conventional ventilation (CV) or high frequency jet or oscillatory ventilation (HFV), used as a rescue measure. We also collected total ventilator days, total ventilator + CPAP days, PMA at last ventilator day, PMA at last CPAP day, and PMAff. Some infants never had an oral feeding attempt due to severity of illness or need for tracheostomy; for this data to contribute to the model variable, these infants were arbitrarily assigned a PMAff of 52w, which was one week beyond the highest PMAff found for all GT infants (51w).
The validation cohort consisted of 187 infants born <30w GA admitted to the NICU at the Medical University of South Carolina from 2017-2018. We recorded the same demographic data for each cohort. In the original cohort, respiratory mode at 32w PMA was the most predictive timepoint for respiratory parameters to predict GT outcome, so respiratory variables were only recorded at 32w PMA in the validation cohort. Any exposure to HFV was also again recorded. In the original model development, we did not look specifically at bronchopulmonary dysplasia (BPD). BPD diagnoses were retroactively assigned to infants in the original cohort using the Shennan et al definition of “oxygen use at 36 weeks PMA.”5 This data is not available in the original model report.
These data were collected with approval of the MUSC Institutional Human Subjects Review Board. Normally distributed continuous data are presented as mean ± standard deviation and P-value derived by t-test; non-normally distributed continuous data are presented as median, interquartile range and compared with Wilcoxon rank sum test; categorical data are presented as number (percent) and tested by chi-squared test. A 2 (GT) by 3 (Fio2) repeated measures ANOVA was utilized to examine Fi02 rates by GT status over the time periods of 30 days, 32 weeks, and 36 weeks. P-values of < 0.05 were considered significant. Pearson’s coefficient was used to test for correlation between respiratory variables and PMAff. Univariate analyses were performed using STATA (State College, TX) and SPSS version 25 (Armonk, NY: IBM Corp). Some figures were created using SPSS version 25.