Establishment And Validation Of Mathematics Model For Estimating The Probability Of Solitary Pulmonary Nodules

ObjectiveTo estimate the probability of malignant of solitary pulmonary nodules(SPNs)by establishing a prediction model which was built by multivariate logistic regression analysis, and to compare our model with the domestic Li Yun model and the foreign models of Mayo model and VA model.MethodsBetween January 2011 and November 2014, 252 patients with solitary pulmonary nodules who had undergone pneumonectomies in the Thoracic Surgery Department of Changhai Hospital and got definite pathological results had been included in our retrospect study. With these cases, we collected clinical and radiologic features including gender, age, symptom, smoking history, history of pulmonary diseases, history of tumor, family history of cancer, the location of lesion, maximum diameter, clear border, smooth border, spiculation, lobulation, pleural indentation,calcification, vessel convergence sign, lucency shadow.209 patients got definite diagnosis before July 18, 2014 was grouped as M0,and it was intended for building the logistic regression model, the cases after that time were grouped as T0,which was planned to test the model and to compare it with the other three classical model. But when the univariate analysis was done, a history of neoplasms within 5 years was founded to be a factor that influenced the malignant probability of the nodules. 19 cases with such histories were founded, with their pathologies show that most of them were metastases(11cases, 57.9%). So the 19 patients who had 5 years history of tumors were abandoned from group M0, and the remaining 190 cases were grouped as M1 to rebuild the model. As time was short,only 43 cases were collected after July 18, 2014. With the group T0,two cases were excluded for they did not meet all the criteria of the other three models at the same time,with the remaining 41 cases defined as group T1. At the same time, 3 cases were ruled out from the group M1 for they did not meet all the criteria of the other three models, and the remaining 187 cases were assigned together with the 41 cases from group T1 to constitute group T2, another testing group with 228 cases.By logistic analysis of the data from group M1, independent factors associated with malignant probability of SPNs were derived and a clinical prediction model wasbuilt, with an appropriate cut-off value also derived. With the data of group T1, this model was verified and was compared with the other three classical models( Mayo model,VA model and Li Yun model). At last, the data from group T2 was also turned to the formulas of the three models.With the calculated probability values for each model from group T1 and T2,the statistical software was used to draw the ROC(Receiver Operating Characteristic)curve, and any two of the AUCs(Area Under The curves) were compared respectively, and the differences were estimated if there were any statistically significant.ResultsSingle factor analysis showed that nine factors differed between benign and malignant nodules, which including age, maximum diameter, smooth border,spiculation, lobulation, pleural indentation, vessel convergence sign, calcification,lucency shadow(P<0.05). While multivariate analysis was done, age, maximum diameter, spiculation, calcification, lucency shadow were the five factors differentiating between BSPLs( benign solitary pulmonary lesions) and MSPLs( malignant solitary pulmonary lesions)(P<0.05)., x=age*0.077+ maximum diameter *0.087+ spiculation *1.366- calcification *2.335+ lucency shadow*1.437-6.222, e was the natural logarithm, cut-off value T = 0.743. When group T1 data was substituted into the formula, the sensitivity was 76.9%, specificity was93.3%,positive likelihood ratio was 11.538, the negative predictive value was 0.247,the positive predictive value was 0.952,the negative predictive value was 0.700.When the data of group T1 was added to the four formulas of Changhai, Li Yun,Mayo and VA model respectively, the corresponding AUCs were 0.910 ± 0.044 、0.794±0.070, 0.700±0.085 and 0.724±0.089, the P value of the Changhai model vs Li Yun model, Changhai model vs Mayo model, or Changhai model vs VA model were both less than 0.05, while the differences between any two of the last three models were not significant. When T2 data was calculated, the AUCs of the last three models were 0.838±0.027, 0.798±0.031, 0.776±0.032, and the difference value of AUC between Li Yun model and VA model was 0.062, P value was 0.018, the P value of the Li Yun model vs Mayo model and Mayo model vs VA model were both greater than 0.5.ConclusionThe patient’s age, maximum diameter, spiculation, calcification, lucency shadow maximum diameter, calcification, lucency shadow are independent predictors of malignant probability. This logistic regression prediction mathematic model has some clinical application value. With patients admitted to the Changhai hospital, our Changhai model seems to work better than the Li Yun model, Mayo model, and VA model.With domestic patients in China, Li Yun model seems more accurate than VA model, with statistical significance, and while compared with Mayo model, the difference is not statistically significant. The difference between Mayo model and VA model does not have statistical sense.