Data Analysis and Power Calculations
Statistical analyses were performed with Stata 16.1 (StataCorp, College
Station, TX, USA). Summary data for continuous variables were reported
as median with interquartile ranges (IQR) and for categorical variables
as numbers with percentages. Continuous data were compared using the
Wilcoxon Rank-Sum test and categorical data were compared by Fisher
exact test. The associations between AVDSf at onset and
echocardiographic findings within the first 24 hours as well as
oxygenation metrics were assessed with univariable and multivariable
logistic regression. Given the relatively small number of patients with
echocardiographic abnormalities, potential confounders were included
into the logistic regression model one variable at a time to not overfit
the model. Additional analyses included the use of AVDSf as a binary
variable based on various cutoffs and the use of RV dysfunction measures
as continuous variables. To test the ability of AVDSf to discriminate
patients with RV systolic dysfunction or PH, area under the receiver
operating curves (AUROC) were calculated. P-values < 0.05 were
considered significant for all analyses.
Initial power analysis using the assumptions of abnormal RV GLS rate of
35% in the population and a difference in AVDSf between groups of 0.15
with a standard deviation of 0.2 resulted in a total population of 65
patients required for the study (80% power, alpha = 0.05).