Statistical analysis
We compared illness differences between diuretic weaning groups, using chi squared or Fisher’s exact tests for differences in proportions, and Kruskal-Wallis tests for differences in medians. To compare our primary outcome of duration of home oxygen therapy, we used Cox survival curves to describe differences between diuretic weaning groups. Differences in secondary outcomes were compared by chi-squared or Fisher’s exact tests or Kruskal-Wallis tests. We similarly assessed whether other illness covariates were associated with differences in duration of home oxygen weaning. To evaluate the association between diuretics and duration of home oxygen therapy adjusted for other illness covariates, we used a series of Cox regression models of time to successful weaning from home oxygen, with failure set as failure to wean by 1 year after discharge. Model was run in steps with key covariate of diuretic management retained at each step: first, an unadjusted comparison of diuretic management; second, adjusted for potential NICU illness covariates that were significantly associated in bivariable analysis with either diuretic weaning plan or duration of home oxygen therapy; third, adjusting for post-NICU significant covariates. Between the second and third steps, variables associated at a p>0.2 were dropped from the model. In the post-NICU-discharge model, emergency and readmission encounters were retained a priori because it was hypothesized that they significantly impacted clinical decision-making about medication management. Variables that were not directly part of the home oxygen weaning process, such as failed room air trials or overnight pulse ox studies, were not included in modeling. A p value of <0.05 was considered statistically significant for all analyses. The institutional Review Board of Children’s Wisconsin approved this study.