Results
A total of 149 infants (63.1% males) were included in the analyses (Table 1).  In our cohort, infants with CNLD requiring home oxygen had a median GA at birth of 26 weeks (IQR 3) and BW 779 grams (IQR 331) with 13 (8.7%) born SGA.  The median length of invasive ventilation during the neonatal period was 17.9 days (IQR 29.4) and mean length of total respiratory support was 73.3 days (SD 28.6) with 53% requiring high frequency oscillation ventilation.  At term CGA, the mean pCO2 was 53.9 mmHg (SD 7.1).  At discharge, the median CGA was 42 weeks (IQR 4.5) with 70.5% of infants requiring 250 mL/min of oxygen flow or less at discharge.  The median CGA of oxygen weaning was 6.8 months CGA (IQR 4.4) with 87.2% of infants able to wean off oxygen by 12 months CGA.
 
Cox proportional hazards regression models for clinical predictors of oxygen weaning at three, six, nine and twelve months CGA with a priori variables of GA, BW, pCO2 at term CGA, duration of invasive ventilation, total duration of respiratory support, length of hospital stay and oxygen flow group at discharge are shown in Table 2.  The models indicated that shorter length of hospital stay was a statistically significant predictor of faster oxygen weaning at CGA 9 months (HR 0.99, 95% CI 0.98–1.00, p=0.02) and at 12 months (HR 0.99, 95% CI 0.98–1.00, p=0.02) (Table 2).  Length of hospital stay was not found to be a significant predictor of oxygen weaning at CGA 3 months (HR 1.01, 95% CI 0.97-1.05, p=0.65) or 6 months (HR 0.99, 95% CI 0.97-1.00, p=0.16).  Higher GA, higher BW, lower pCO2 at term CGA, fewer invasive ventilation days and fewer total days on respiratory support were not significant predictors of faster oxygen weaning at any time point (Table 2).
 
Our multinomial logistic regression (Table 3) indicated that infants has higher odds of being discharged with 200-250 mL/min (OR 1.099, 95% CI 1.02-1.19, p=0.02) or 251-750 mL/min (OR 1.13, 95% CI 1.04-1.23, p=0.004) home oxygen relative to ≤ 125 mL/min with increasing measured pCO2 at term CGA when all other variables were held constant.  With each week increase in CGA at hospital discharge, there were higher odds of being discharged on 200-250 mL/min (OR 1.34, 95% CI 1.09-1.63, p= 0.005) or 251-750 mL/min (OR 1.51, 95% CI 1.21-1.88, p<0.001) relative to ≤ 125 mL/min when all other variables were held constant.  With every day increase in length of hospital stay, infants had lower odds of being discharged on 200-250 mL/min (OR 0.95, 95% CI 0.91-0.99, p=0.02) or 251-750mL/min (OR 0.94, 95% CI 0.90-0.99, p=0.01) relative to ≤ 125 mL/min when all other variables were held constant.  Our results also suggest that infants born to mothers with PROM had lower odds of being discharged on 200-250 mL/min home oxygen (OR 0.36, 95% CI 0.14-0.91, p=0.032) compared to mothers without PROM when all other variables were held constant (Table 3).
 
For our model, accuracy was around 55% (CI 46-63%) with performance of the model using diagnostic measures outline in Appendix 2.  Sample sizes were low in some groups which is reflected by the accuracy of the model.
 
We found significant associations of oxygen flow rate at discharge with respiratory related hospital admissions (p<0.01) and PICU admissions (p=0.05) up to two years of chronological age (Table 4).   We estimated that infants discharged on 251-750 mL/min of home oxygen had 6.08 times higher odds (95% CI, 2.20-18.48, p<0.001) of being admitted (at least once) with a respiratory related illness within the first two years of chronological age (Appendix 3).