Plasma metabolomics enhances risk prediction for lung cancer: a population-based validation study

Metabolic biomarkers offer promising opportunities for lung cancer risk assessment and early detection. This study aimed to identify metabolites with both screening and early diagnostic value, and to develop and externally validate a metabolite-based risk prediction model for lung cancer. We employed a hospital-based case-control design combined with a population-based nested case-control study. In the first stage, 210 lung cancer patients and 210 matched healthy controls were included. In the second stage, we included 142 incident lung cancer cases with pre-diagnostic blood samples and 264 cancer-free participants from the Cancer Screening Program in Urban China. Multivariable logistic regression was used to construct risk models based on epidemiological risk factors alone, and in a combination of metabolites. Five overlapped differential metabolites were identified in both prevalent cases and pre-diagnostic samples: decanoylcarnitine, histidine, tryptophan, L-carnitine, and octanoylcarnitine. The area under the curve (AUC) of the five-metabolite-incorporated risk model was 0.878 (95% CI: 0.832, 0.925), outperforming models based solely on epidemiological risk factors. Model performance was consistently validated internally (AUC 0.858; 95% CI: 0.808, 0.907) and externally (AUC 0.770; 95% CI: 0.702, 0.837). Logistic regression demonstrated the strongest predictive ability compared with machine-learning approaches. Our findings provide insights that incorporating metabolomic profiles into risk assessment may significantly enhance the identification of high-risk individuals and facilitate precise screening and early detection.