Establishment and validation of mathematics model for differentiating benign and malignant solitary pulmonary nodules
10.3724/SP.J.1008.2015.00407
- Author:
Juan YANG
1
Author Information
1. Department of Respiratory Medicine, Changhai Hospital, Second Military Medical University
- Publication Type:Journal Article
- Keywords:
Clinical pathology;
Logistic models;
Lung neoplasms;
Solitary pulmonary nodule
- From:
Academic Journal of Second Military Medical University
2014;36(4):407-412
- CountryChina
- Language:Chinese
-
Abstract:
Objective To establish a prediction model by multivariate logistic regression analysis for estimating the malignant probability of solitary pulmonary nodules (SPNs), and to compare our model with other models. Methods From January 2011 to November 2014, totally 252 patients with SPNs who had undergone pneumonectomies in Thoracic Surgery Department of Changhai Hospital and been confirmed with definite pathological results were included in this retrospective study. The gender, age, symptom, smoking history, history of pulmonary diseases, history of tumor, family history of tumor, the location of lesion, maximum diameter, clear border, smooth border, spiculation, lobulation, spinous protuberant sign, pleural indentation, calcification, vessel convergence sign, and lucency shadow were all reviewed. Eighty-three cases were designated as validation group (group B), and the remaining 169 cases were taken as the modeling group (group A). Six cases were excluded from group B to meet all the inclusion and exclusion criteria of the other three models, and the remaining 77 cases constituted group C. Logistic analysis identified five independent factors associated with malignant probability of SPNs from group A and a clinical prediction model was built. With the data of group B and group C, this model was verified and was compared with the other three classical models. Results The age, history of tumor, maximum diameter, calcification, and lucency shadow were the five factors identified for differentiating benign and malignant SPNs (P<0.05). When group B data was substituted into the established formula, the area under curve (AUC) of the ROC was 0.905±0.036, sensitivity was 79.3%, specificity was 84.0%, positive likelihood ratio was 4.957, negative likelihood ratio was 0.246, positive predictive value was 0.920, and negative predictive value was 0.636. When the data of group C were fed to the four formulas of Changhai, Li Yun, Mayo and VA models, the corresponding AUCs were 0.893±0.040, 0.817±0.056, 0.804±0.050, and 0.780±0.057, respectively. Conclusion The patient's age, history of tumor, maximum diameter, calcification, and lucency shadow are the independent predictors of malignant probability of SPNs. This logistic regression prediction mathematic model is of clinical application value. For patients in this study, our Changhai model seems to work better than the Li Yun, Mayo,and VA model.
