Clinical verification and application of mathematical models for predicting the probability of malignant or benign in patients with solitary pulmonary nodules / 中华胸心血管外科杂志

ABSTRACT

Objective Based on the mathematical models established in Department of Thoracic Surgery of Peking University People's Hospital for predicting malignant probability for solitary pulmonary nodules ( SPN),another continuous 145 patients with SPN were assessed to verify the accuracy of the model comparing with foreign models (Mayo model and VA model).Methods A retrospective cohort study in our institution included 145 patients with definite pathological diagnosis of SPN from Oct 2009 to Aug 2011,72 males and 73 females,average age (59.4 ± 12.2 ) years old.Clinical data included age,gender,course of disease,symptoms,history and quantity of smoking,time of smoking cessation,history of tumor,family history of tumor,tumor site,diameter,calcification,speculation,border,lobulation,traction of pleural,vascular convergence sign,and cavity.These raw data were incorporated into our model,Mayo model and VA model,the probability of malignant in every patient was calculated separately according to methods described before.The sensitivity and specificity of these 3 models were evaluated then.Afterwards,calibration of the 3 models was assessed by the Hosmer-Lemeshow (H-L) test.Discrimination was tested by calculating the area under curve ( AUC ) after the receiver operating characteristic (ROC) curve was drawn.Results 32.4％ (47 in 145 patients) of the nodules were malignant,and 67.6％ (98 in 145 patients) were benign in this group.Verified the accuracy of our model with sensitivity of 94.9％,specificity of 66.0％,positive predictive value of 85.3％ and negative predictive value of 86.1％.The H-L test showed good fitting in all models ( P ＞0.05 ).The AUC for our model was 0.874 ±0.035,and 0.784 ± 0.041 in Mayo model (P =0.004 compared to our model),0.754 ± 0.041 in VA model (P =0.002 compare to our model).And,there was not significant statistical difference between Mayo model and VA model (P ＞0.05 ).Our model has the best precision indexed by AUC,which were statistically significant differential compared with Mayo model and VA model.Conclusion The model established by our center has superior value than foreign counterparts in predicting the probability of malignant or benign in patients with SPN.