Statistical analyses
Groups, defined as HFNC failure or success, were compared using Student or Wilcoxon Mann-Whitney test for continuous variables and Chi-square or Fisher test for categorical ones.
The accuracy of ROX index and physiological variables to predict HFNC failure was assessed using receiver operating curve (ROC) analysis. The area under the curve (AUC) was calculated by the Hanley method and compared to the value 0.5 using Wilcoxon’s W statistic. For the analyses at H1 and for changes during the first hour of HFNC delivery, all HFNC failure patients at H1 or before were excluded.
Statistical tests were performed 2-tailed and p-values <0.05 were considered significant test results. Statistical analyses were conducted with SAS (version 9.4, SAS Institute, Cary, NC) and R software (R 2.3.4 for Windows). Patients’ characteristics were presented using mean and standard deviation (SD) for the quantitative variables and frequencies with proportions for categorical ones.