Proteomics out-performs existing prediction tools for lung cancer risk discrimination


Lung cancer remains the leading cause of cancer death globally, with late diagnosis of advanced stage cancer contributing to the poor prognosis. A group from the International Agency for Research on Cancer in Lyon has developed a proteomics risk model for lung cancer and compared its performance against existing prediction tools. Using four Olink® Target 96 panels, blood proteomics were measured in subjects from 6 independent cohorts that were part of the INTEGRAL (Integrative Analysis of Lung Cancer Etiology and Risk) project. Protein levels were compared between 624 patients who developed lung cancer up to 3 years after baseline sampling during the INTEGRAL project and 624 smoking-matched controls and used to identify a proteomics-derived risk model that could be evaluated in relation to an existing smoking-based clinical risk model (PLCOm2012) and a commercially available autoantibody biomarker test ((EarlyCDT-Lung).


A total of 22 proteins were associated with lung cancer risk after correction for multiple testing in the development dataset, with 8 of these (CXCL17, IGFBP-1, CXCL13, CASP-8, CDCP1, IL6, MMP12, CEACAM5) selected by multivariable LASSO logistic regression in at least 50% of the training sets. Further analysis in the validation sample set showed that the protein-based model had a risk prediction AUC=0.75, compared to 0.64 for the PLCOm2012 model. The EarlyCDT-Lung commercial autoantibody test had a sensitivity of just 14% and a specificity of 86% for incident lung cancer. At the same specificity of 86%, the sensitivity for the protein-based risk model was estimated at 49%, compared to 30% for the PLCOm2012 clinical model. The conclusion was that circulating proteins showed promise in predicting incident lung cancer, out-performing both the standard risk prediction model and the commercialized test.

One additional observation of note from this study was the identification of CEACAM5 as the most predictive individual protein in the model – this marker has been shown to predict multiple types of metastatic cancer several years prior to clinical diagnosis in a previous Olink-based study (Magis et al., 2020, Scientific Reports, DOI: 10.1038/s41598-020-73451-z). A related article from the INTEGRAL project has also recently reported that estimates of the stage of undiagnosed lung cancer correlating to the baseline samples taken for the Olink analysis suggest that a significant majority of patients (78%) were likely to have had stage 2 or earlier cancer. This could indicate that the protein biomarkers identified have the potential to detect many cases of lung cancer at a more readily curable stage (Lung Cancer Cohort Consortium (LC3), 2023, Nature Communications, DOI: 10.1038/s41467-023-37979-8).



Feng X, Wu WY, Onwuka JU, et al. Lung cancer risk discrimination of prediagnostic proteomics measurements compared with existing prediction tools. (2023) Journal of the National Cancer Institute, DOI: 10.1093/jnci/djad071

The protein-based model performed well in the validation sample and outperformed the EarlyCDT-Lung and PLCOm2012 model in relevant strata

Feng et al. (2023)

Peer-reviewed publications citing the use of Olink panels

Olink’s Proximity Extension Assay (PEA) technology has been used for protein biomarker discovery and analysis across a very broad range of disease areas and applications, providing actionable insights into disease biology and helping to drive future development of new and better therapeutics. There are now well over 1200 publications citing the use of our assays and the list is growing rapidly. Please visit our library of publications to see some of the extraordinary work produced by Olink customers.