Next-generation proteomics improves lung cancer risk prediction

Screening heavy smokers by low-dose computed tomography (LDCT) can reduce lung cancer (LC) mortality, but defining the population that benefits most, a prerequisite for cost-effective screening, is challenging. In order to contribute to a more nuanced risk stratification of high-risk target populations, we developed and validated a blood-based protein marker model for LC. A two-stage design was implemented in this study, and the derivation set comprised 18 868 participants from the UK Biobank, which included 200 incident LC cases identified at 6 years of follow-up. The independent validation set included 101 LC cases identified at 6 years of follow-up. A total of 2025 protein markers measured by proximity extension assays available for both datasets were used for analysis. A risk prediction algorithm by least absolute shrinkage and selection operator regression with bootstrap method was developed in the derivation set and then externally evaluated in the independent validation set. The risk discriminatory performance of the protein marker model was compared with the established PLCO_m2012 model, USPSTF 2020 guidelines and trial criteria used in different LDCT trials. The protein marker model comprising of four protein biomarkers-CEACAM5, CXCL17, MMP12, and WFDC2-outperformed the PLCO_m2012 model, and the areas under the receiver operating curve (AUCs) for the protein marker model in the derivation and validation sets were 0.814 [95% confidence interval (95% CI), 0.785-0.843] and 0.814 (95% CI, 0.756-0.873), respectively. The addition of the protein marker model to the PLCO_m2012 model increased the AUCs up to 0.056 and 0.057 and yielded up to 16 and 12 percentage points higher sensitivities to identify future LC cases compared to the LDCT trial criteria, in the derivation and validation sets, respectively. The protein marker model improves the selection of high LC risk individuals for LDCT screening and thereby enhances screening efficacy.