Table 2.
Results of ROC analysis (Malignant versus Benign).
| AUC (95% CI) | P value | Cut-off value | Sensitivity (%) | Specificity (%) | |
|---|---|---|---|---|---|
| Training set | |||||
| MTD | 0·601 (0·506–0·696) | 07 | 1·9 | 33·3 | 91·9 |
| FR+-CTC | 0·781 (0·698–0·864) | <001 | 8·3 | 78·6 | 78·4 |
| CEA | 0·637 (0·543–0·731) | 01 | 1·81 | 38·5 | 91·9 |
| CYFRA21-1 | 0·694 (0·598–0·791) | <001 | 1·15 | 76·1 | 62·2 |
| NSE | 0·648 (0·543–0·752) | 007 | 10·84 | 73·5 | 56·8 |
| SCC | 0·562 (0·457–0·667) | 0·26 | 0·91 | 50·4 | 70·3 |
| CEA + CYFRA21-1 + NSE | 0·742 (0·657–0·828) | <001 | 0·700 | 70·9 | 70·3 |
| FR+-CTC + CEA + CYFRA21-1 + NSE | 0·864 (0·807–0·921) | <001 | 0·825 | 65·8 | 94·6 |
| Validation set | |||||
| MTD | 0·552 (0·386–0·719) | 53 | 1·9 | 73·1 | 43·8 |
| FR+-CTC | 0·792 (0·668–0·917) | <001 | 8·3 | 82·7 | 68·8 |
| CEA | 0·576 (0·425–0·726) | 36 | 1·81 | 40·4 | 62·5 |
| CYFRA21-1 | 0·536 (0·387–0·686) | 67 | 1·15 | 67·3 | 37·5 |
| NSE | 0·614 (0·444–0·784) | 17 | 10·84 | 73·1 | 43·8 |
| SCC | 0·509 (0·338–0·680) | 91 | 0·91 | 46·2 | 50·0 |
| CEA + CYFRA21-1 + NSE | 0·618 (0·463–0·773) | 16 | 0·700 | 69·2 | 37·5 |
| FR+-CTC + CEA + CYFRA21-1 + NSE | 0·802 (0·673–0·930) | <001 | 0·825 | 55·8 | 81·3 |
Only patients who were randomly selected to receive serum biomarker tests (117 lung cancer and 37 benign lung disease patients in the training set, 52 lung cancer and 16 benign lung disease patients in the validation set) were included in this ROC analysis. The logistic equations and cut-off values established from the training set data were confirmed using the validation set data. The equations are as follow: Probability of malignancy = eα/(1 + eα); For “CEA + CYFRA21-1 + NSE”, α = 0·602*CEA + 0·573*CYFRA21-1 + 0·041*NSE – 1·168; For “FR+-CTC + CEA + CYFRA21-1 + NSE”, α = 0·388*FR+-CTC + 0·736*CEA + 0·633*CYFRA21-1 + 0·004*NSE – 4·346.