2.4 Statistical Analysis
The categorical and continuous variables were compared using the χ2 and
Student’s t‐test, respectively. Wilcoxon’s rank-sum test was used for
comparisons of lung density indices in CT scans with and without BO.
Correlations between the various quantitative lung density indices and
PFT parameters were compared using Spearman’s rho (ρ), due to the
heteroscedastic nature of the data. We also tested the association
between LDIs and PFTs in linear regression models, with adjustment for
age and sex. In multivariate logistic regression, we analyzed the odds
ratio (OR) of the diagnosis of BO by lung density and PFT parameters.
Stepwise backward elimination and all subset regressions were performed
to select the final model in multivariate analysis. The receiver
operating characteristic (ROC) method was performed to evaluate the
utility of different lung density parameters for predicting
physician-diagnosed BO. Areas under the curve (AUCs) and optimal cut‐off
points based on maximizing the sum of sensitivity and specificity were
calculated for each densitometry parameter. P values of <0.05
were considered statistically significant; the statistical analyses were
performed using R Version 4.0.5
(https://cran.r-project.org/web/packages/maxstat/index.html). This study
was approved by the Institutional Review Board of Seoul St. Mary’s
hospital.