Joint prediction of solitary pulmonary module malignant probability based on logistic regression and malignant tendency comprehensive score.

Abstract

Purpose: We analyzed the relationship between clinical data, tumor markers, chest high-resolution CT(HRCT) and pathology in patients with solitary pulmonary nodules (SPN) and explored the joint discrimination scheme to improve the accuracy of noninvasive diagnosis.

Methods: 242 SPNs with the largest diameter D<2cmwere divided into training set (161 cases) and test set (81 cases). We screened the risk factors by single factor analysis. Then, we established the prediction equation model (PE model) based on logistic regression and malignant tendency comprehensive score model (MTCS model) based on the evaluation criteria of SPN. The weight of the two sub models was used to determine the joint evaluation model (JE model).

Results: Age, CEA content, maximum diameter, pleural adhesions, spicule sign, and ground glass component were independent factors of malignant prediction (p<0.05) recorded as x1~x6, and PE model was established as P1=ex/(1+ex),x=0.052x1+0.0327x2+0.212x3+1.849x4+ 1.066x5+1.769x6-7.582.According to the different performance of different manifestations of the corresponding score, we could get each score S of SPN. The MTCS model was S/8.5. The JE model was P=0.76P1+0.24S/8.5. The results of test set showed the AUC values of JE, PE, MTCS, Mayo, VA and Li Yun model for D≤2cm SPN were 0.930(95% CI:0.877-0.983), 0.922(95% CI:0.870-0.974), 0.900(95% CI:0.879-0.921), 0.782(95% CI:0.749-0.815), 0.744(95% CI:0.731-0.756) and 0.801(95% CI:0.739-0.863). The sensitivity of JE, PE, MTCS model were 87.2%, 79.2%, 73.3%, the specificity was 90.1%, 89.2%, 82.2%, and the accuracy was 89.9%, 85.5%, 81.2%.

Conclusions: The joint evaluation model has better diagnostic efficiency and can provide reference for the diagnosis of SPN with D≤2cm.