Abstract
OBJECTIVE: Low dose computed tomography (LDCT) screening has been proven to be effective in reducing lung cancer mortality, but the ensuing high false-positive and overdiagnosis rates shackle the effectiveness of lung cancer screening (LCS) in China. Nodule malignancy prediction models may be an applicable solution. METHODS: We conducted a prospective cohort study to develop and internally validate the model using data from the ongoing Henan province Cancer Screening Program in Urban China (CanSPUC). From 2013 to 2021, 23,031 heavy smokers underwent baseline screening with LDCT; 2553 participants were diagnosed with pulmonary nodules. Detailed questionnaire, physical assessment and follow-up were completed for all participants. Multivariable Cox proportional risk regression models were used to identify and integrate key prognostic factors for the development of a nomogram model. Data from the National Lung Screening Trial (NLST) were utilized for external validation. RESULTS: A total of 111 lung cancer cases with a median follow-up duration of 3.7 years occurred in the Henan CanSPUC. Age, gender, physical activity, consumption of pickled food, history of silicosis or pneumoconiosis, nodule type, size, calcification, and pleural retraction sign were included into the model. The AUC was 0.855, 0.844, and 0.863 for the 1-, 3- and 5-year lung cancer risk in the training set, respectively. Compared with Mayo model, VA model, PKU model, and Brock model, the Henan CanSPUC model yield statistically better discriminatory performance (all P values < 0.05). The model calibrated well across the deciles of predicted risk in both the overall population and all subgroups. The model demonstrated good calibration and discrimination in the internal validation cohort, while the external validation cohort showed lower predictive performance, indicating that further external validation is needed. CONCLUSIONS: The model developed and validated in this study may be used to estimate the probability of lung cancer in nodules detected at baseline LDCT, allowing more efficient risk-adapted follow-up in population-based LCS programs. However, further external validation in broader and more diverse populations is warranted.