Evaluation of Four Predictive Models for Identifying Malignancy of Solitary Pulmonary Nodules in Health Check-up Population

Abstract

Objective To compare and validate the efficiency of four models predicting the malignancy of solitary pulmonary nodules (SPN). Methods Patients diagnosed with SPN during health check-up were selected as the research subjects. Risk assessment was conducted using four predictive models. Outcomes were obtained through prospective follow-up. Statistical description and univariate analysis were performed for all risk factors of the four models. ROC curve was applied to compare the efficiency of the four predictive models. Results A total of 479 cases were included in this study. Among these patients, 82 were diagnosed with lung tumor, and the malignant rate was 17.12%. Age, nodule diameter, smoking, family history of tumor, history of extrapulmonary tumor ≥5 years, upper lobe site, unclear boundary, and spiculation rates were higher in the malignancy group than those in the benign group (P < 0.05). The efficiency of Brock model was the best. Its AUC was 0.833, sensitivity was 80.49%, and specificity was 74.31%. Its Youden index, positive likelihood ratio, positive predictive value, and negative predictive value were the highest, and its negative likelihood ratio was the lowest. The AUC, sensitivity, and specificity of Mayo model were 0.815, 81.71%, and 67.51%, respectively; those of PKUPH model were 0.754, 69.51%, and 73.55%, respectively; and those of VA model were 0.738, 68.29%, and 67.55%, respectively. Conclusion The Brock model might be the most appropriate predictive model for the risk assessment of SPN among the health check-up population, and the VA model is the worst. The combination of Brock, Mayo, and PKUPH models requires further study.

Keywords

• predictive models for lung cancer
• spn
• health check-up
• roc curve
• prospective study