3.3 Logistic Regression and ROC Analysis
Multivariate logistic regression showed the OR of the LDIs for the BO
diagnosis (Table 3). PFT parameters related to BO diagnosis were low
FEV1 (OR = 0.89, p < .001), FEV1/FVC (OR = 0.89, p
< .001), FEF25-75 (OR = 0.94, p < .001), and high RV
(OR = 1.03, p < .001). MLDD (OR = 0.98, p < .001),
E/I MLD (OR = 1.39, p < .001), E/I Volume (OR = 1.11, p
< .001), E900 (OR = 1.42, p = 0.036), and E600 (OR = 0.90, p
< .001) were statistically significant densitometry parameters
for BO diagnosis.
In the ROC analysis, the optimal cut-off values of each MLD and PFT
parameter for BO diagnosis were obtained (Figure 3). The sensitivity,
specificity, positive predictive value, negative predictive value, and
AUC for each cut-off value of LDIs and PFT parameters are presented in
Table 4. The parameters showing high AUC for BO diagnosis were FEF 25-75
(cut-off = 56.0%; AUC = 0.961), FEV1 (cut-off 81%; AUC = 0.931),
FEV1/FVC (cut-off = 80%, AUC = 0.897) in PFTs, and E900 (cut-off 1.4%;
AUC = 0.920), E/I MLD (cut-off 86.9%; AUC = 0.887), and MLDD (109 HU;
AUC = 0.867) in MLD parameters, respectively.