Endobronchial Ultrasound Combined with Clinical Data for Predicting Malignant Peripheral Pulmonary Lesions
2020
Introduction This study was to develop a simple model for predicting malignancy of peripheral pulmonary lesions (PPLs) based on endobronchial ultrasonography (EBUS) and clinical findings. Methods Patients who had EBUS for PPLs were analyzed and compared on the EBUS imaging characteristics and clinical data. The malignancy prediction model was established by the logistic equation of probability of malignant PPL based on the data of 135 patients. The model was tested on an additional 50 patients for efficiency. Results Among 135 prospectively enrolled patients, 77 (57%) patients had malignant and 58 (43%) had benign lesions with the size of 36.5±19.9 mm. Univariate analysis demonstrated a significant (P 0.05) difference in age (50 years as the cutoff value) and history of extra-thoracic malignancies. Logistic analysis of multiple factors showed that smoking history, serum CEA, borderline, air bronchogram, heterogeneous echo, and anechoic areas were significant (P<0.02) risk factors for malignant lesions. The malignancy prediction model was established by the logistic equation of probability of malignant PPL (P) = l/[l+e-Z], where Z=-2.986+1.993X1+2.293X2+l.552X3+1.616X4-2.011X5+1.718X6, e is the base of the natural logarithm, X1 is the smoking history, X2 is the serum CEA, X3 is the borderline, X4 is the heterogenicity, X5 is the air bronchogram, and X6 is the anechoic area. The receiver operating characteristic curve had an area under the curve (AUC) of 0.926 (95% confidence interval: 0.883-0.969). The sensitivity, specificity, and accuracy were 88.2% (30/34), 75.0% (12/16), and 92.0% (46/50), respectively, for the logistic equation to predict the malignancy. Conclusion Endobronchial ultrasonography is a safe and practical method, and the model combining EBUS and clinical data can accurately predict the malignancy of peripheral pulmonary lesions.
- Correction
- Source
- Cite
- Save
- Machine Reading By IdeaReader
22
References
1
Citations
NaN
KQI