Statistical analysis
We compared the distribution of pre-discharge pCO2 between the NICU and referral groups. We then compared differences between pre-discharge pCO2 by NICU variables; we repeated these comparisons using 36-week pCO2 for the NICU group. Similarly, we compared differences between pCO2 and outcomes. For all bivariable comparisons, we used sign-rank or Kruskal-Wallis tests, as appropriate. For the primary outcome of respiratory readmissions, we evaluated the association between pCO2 and readmission before and after adjustment for other NICU illness covariates. We chose illness covariates for inclusion in the model based on both bivariable analysis and the clinical possibility that they could impact the association between pCO2 and readmission; these included gestational age, BPD severity, ductus arteriosus ligation, and diuretic use 12,13. To further describe our secondary outcome of duration of home oxygen use, we considered both corrected gestational age at discontinuation of home oxygen in categorized groups, as well as a survival curve of weeks post-NICU discharge on home oxygen. For all analyses, p value of <0.05 was considered statistically significant. This study was approved by the Institutional Review Board of Children’s Wisconsin.