Background?Lung cancer is the leading cause of cancer morbidity and mortality in China, accounting for over one-third of global cases and deaths. Despite evidence that LDCT screening significantly reduces mortality among heavy smokers in Western studies, there is a high proportion of lung cancer cases among non-smokers (LCINS) in China, with limited data on screening effectiveness for this group. Additionally, while risk prediction models, tumor biomarkers, and polygenic risk scores (PRS) have the potential to enhance risk stratification, they lack validation in large-scale screening settings in China.
Objectives?This study aims to establish a large-scale lung cancer screening cohort in Wenling City, Zhejiang Province, to assess screening outcomes and compare the effectiveness of lung cancer-specific macroscopic risk prediction models, tumor biomarkers, and polygenic risk scores in stratifying risk among high-risk smokers and non-smokers.
Methods?A population-based LDCT screening cohort for lung cancer was established in Wenling City, Zhejiang Province, enrolling 28,910 high-risk residents aged 50–74 years between 2019 and 2021. Participants were classified as smokers (n=15,774) or non-smokers (n=13,136). Screening outcomes, lung cancer incidence, stage distribution, and interval cancers were assessed with follow-up through December 31, 2023 (mean follow-up 3.32 years). Seventeen published lung cancer risk prediction models, four tumor markers (CEA, CYFRA21-1, NSE, SCC-Ag), and ten PRSs were evaluated. Model discrimination was assessed using AUC, with incremental value quantified by NRI, IDI, and decision curve analysis.
Results?A total of 560 incident lung cancers were identified (overall incidence rate 1.65%), with 67.3% diagnosed at an early stage. Smokers had a higher incidence rate than non-smokers (1.88% vs. 1.39%), but also experienced more interval cancers and longer diagnostic delays. Lung cancer in non-smokers was almost exclusively adenocarcinoma, with a high proportion of early-stage cases and adenocarcinoma in situ, suggesting potential overdiagnosis. Among smokers, the macroscopic LCRS model had the highest performance (AUC 0.663), while the best-performing model in non-smokers was the TNSF-NG model (AUC 0.585). CEA was the only tumor marker linked to lung cancer risk in both groups, with CYFRA21-1 and SCC-Ag associated only in smokers. Combining CEA, CYFRA21-1, and SCC-Ag improved discrimination in smokers (AUC 0.628) over CEA alone (AUC 0.608), while CEA showed limited discrimination in non-smokers (AUC 0.559). In smokers, the best-performing PRS was PRS-33 (AUC 0.546), while the PRS-19 model achieved the highest discrimination in non-smokers (AUC 0.627). Comprehensive models significantly improved discrimination in both groups, reaching AUCs of 0.690 (NRI 35.08%) in smokers and 0.662 (NRI 39.00%) in non-smokers.
Conclusions:Large-scale LDCT screening in China enables effective early detection of lung cancer but requires differentiated strategies. Annual screening is essential for smokers, including those with negative baseline results, while extended intervals may be appropriate for non-smokers to reduce overdiagnosis. Macroscopic risk models and tumor markers are more informative in smokers, whereas PRS provides superior risk discrimination in non-smokers, with comprehensive models offering further gains. These findings support individualized, risk-adapted lung cancer screening strategies tailored to the Chinese population.

Flow diagram