Landscape of volatile organic compounds for differential diagnosis, pathological typing, and severity prediction in lung cancer: a large population-based prospective study

BACKGROUND: In this study, we have developed the largest sample size model to achieve non-invasive, convenient and low-cost lung cancer diagnosis and discrimination using Volatile organic compounds (VOCs) analysis. METHODS: We recruited patients with pulmonary nodules and healthy individuals from three hospitals in Sichuan Province, China, for prospective exhaled air acquisition and subsequent analyses. Participants were 50 to 74 years of age and nodules confirmed on low-dose computer tomography. We developed machine learning models based on exhaled breath composition to distinguish lung cancer patients from health individuals, benign and malignant lung nodules, stage and pathological type of lung cancer, and prognostic risk models. The main outcome was to propose meaningful exhaled breath VOCs for different diagnosis and treatment purposes, and finally to evaluate the performance of the model. Participants were recruited from the ResMan registry (registration number: ChiCTR-DOD-17011134) and had completed 5 years of follow-up. FINDINGS: O, cyclopentane, AMS, and methyl sulfide). The prediction model exhibited stable and powerful predictive prognostic capacity. INTERPRETATION: Our data can differentiate between lung cancer and healthy individuals, distinguish between benign and malignant lung nodules, as well as perform staging, typing, and prognosis analysis for lung cancer. This provides new perspectives and evidence for the application of exhaled gas composition analysis in lung cancer diagnosis.