Efficacy of three lung cancer prediction models in diagnosing benign and malignant pulmonary nodules.

Background: The predictive models for malignant lung nodules have been developed, but need further validation and optimization for broader clinical use. This study aimed to compare the diagnostic efficacy of the Mayo model, Peking University People's Hospital (PKUPH) model, and the lung cancer biomarker panel (LCBP) model in distinguishing between benign and malignant pulmonary nodules, providing valuable clinical research data for the early diagnosis of lung cancer. Methods: Clinical and imaging data of patients diagnosed with pulmonary nodules at Meizhou People's Hospital from March 2021 through January 2023 were collected. Data from patients with benign pulmonary nodules during the same period, who served as negative referents, were also gathered. The Mayo model, PKUPH model, and LCBP model were used to clinically validate lung cancer prediction rates. The receiver operating characteristic (ROC) curves and statistical significance comparing the areas under the curve (AUCs) for each model were evaluated. Results: A total of 428 patients were included: 160 females and 268 males. The noncancer group included 218 cases (50.93%), and the cancer group included 210 cases (49.07%). The AUC values of the three models were as follows: Mayo model, 0.783; PKUPH model, 0.726; and LCBP model, 0.759. (I) For the Mayo model, at the maximum Youden index, the concordance rate was 74.3%, the sensitivity 85.71%, the specificity 63.30%, the positive predictive value 69.23%, the negative predictive value 82.14%, the positive likelihood ratio 2.335, and the negative likelihood ratio 0.226. (II) For the PKUPH model, at the maximum Youden index, the concordance rate was 70.3%, the sensitivity 84.29%, the specificity 56.88%, the positive predictive value 65.31%, the negative predictive value 78.98%, the positive likelihood ratio 1.955, and the negative likelihood ratio 0.276. (III) For the LCBP model, at the maximum Youden index, the concordance rate was 75.0%, the sensitivity 72.38%, the specificity 77.52%, the positive predictive value 75.62%, the negative predictive value 74.45%, the positive likelihood ratio 3.220, and the negative likelihood ratio 0.356. Conclusions: All three predictive models exhibit clinical applicability, with minimal differences in diagnostic efficacy. The LCBP model outperformed both the Mayo and PKUPH models in diagnostic performance, showing greater diagnostic value for the Chinese population. However, there is still room for optimization in each model.