Plasma proteomic profiles for early detection and risk stratification of non‐small cell lung carcinoma: A prospective cohort study with 52,913 participants

Early detection of non-small cell lung cancer (NSCLC) can improve survival rates, and plasma proteomics may provide effective tools for risk prediction. The population for this study included 52,913 participants and 2911 plasma proteomics from UK Biobank. The cohort was divided into discovery and validation cohorts based on their countries. Cox regression, XGBoost, and SHAP analysis were used to identify key NSCLC-associated proteins. Machine learning (ML) models were developed and validated across different timeframes. Risk stratification was performed using protein levels. Temporal change analysis and two-sample Mendelian Randomization (MR) were conducted to assess the early prediction ability and causal relationships, respectively. Twenty-five proteins were significantly associated with NSCLC. ML identified CXCL17, CEACAM5, and WFDC2 as having the highest predictive power. The three-protein panel plus epidemiological indicators exhibited superior performance in 5- and 10-year predictions, achieving AUROC of 0.904 (95%CI: 0.839-0.968) and 0.873 (95%CI: 0.815-0.931), respectively. Risk stratification identified a high-risk group with a 9.18-fold higher risk than the general population and a 16.75-fold higher risk than the low-risk group, respectively. Temporal change analysis revealed that protein expression levels in cases were globally higher than in controls up to 10 years before diagnosis. MR implied a suggestive causal relationship between CXCL17 and NSCLC. Our findings suggest three plasma proteins possess robust predictive capabilities for NSCLC, allowing for predictions up to 10 years in advance. Incorporating these biomarkers into risk models enhances early detection, providing a foundation for targeted screening and precision medicine in NSCLC.