Integrating Plasma Proteomics and Polygenic Risk Scores for the Prediction of Lung Cancer Risk

Multimodal diagnostic strategies and risk stratification are essential for improving early lung cancer detection, clinical outcomes, and personalized interventions. To develop a predictive model for lung cancer incidence by integrating plasma proteomics, genetic factors, and demographic data. We analyzed data from the UK Biobank, which included 47,655 adults without a history of lung cancer at baseline, with 641 incident cases and a median follow-up duration of 13.87 years. COX proportional hazards modeling with SHapley Additive exPlanations (SHAP) sequential feature selection was used to identify significant plasma proteins associated with lung cancer. An XGBoost-based survival model was constructed and evaluated to assess its predictive utility. Proteomic risk score (ProRS), demographic, and Combined risk scores were developed to evaluate their predictive capabilities across genetic risk strata. Among 2911 plasma proteins, we identified 15 significant plasma proteins associated with the development of lung cancer. The XGBoost survival model demonstrated promising predictive performance, achieving a C-statistics of 0.803. Risk stratification revealed that high-risk groups had up to a 101-fold increased risk of lung cancer compared to low-risk groups. We analyzed expression patterns of 15 carcinogenesis-associated proteins during the 15-year period preceding diagnosis. TNR and CLEC3B were identified as the earliest dysregulated proteins, with expression levels consistently downregulated beginning as early as 14 years before diagnosis. GDF15 exhibited sustained and pronounced upregulation throughout the entire 14-year observation window. Integrating plasma proteomics, polygenic risk scores, and demographic factors improves personalized lung cancer risk stratification, offering potential for refined screening strategies.