Exhaled-Breath Volatile Fingerprints Detect Early-Stage COPD and Stratify Disease Severity

Chronic obstructive pulmonary disease (COPD) is a leading cause of death worldwide, yet most cases remain undiagnosed until substantial, irreversible airflow limitation has occurred because case-finding depends on spirometry. Here we show that rapid, non-invasive profiling of exhaled volatile organic compounds (VOCs) can detect COPD, including early-stage disease, and stratify severity. In a prospective, two-centre, observational case-control study of 825 breath samples, we used proton-transfer-reaction time-of-flight mass spectrometry to derive a compact, interpretable 16-compound VOC “fingerprint” and trained eight classifiers. A k-nearest neighbours model distinguished COPD from controls with an area under the receiver operating characteristic curve (AUC) of 0.955 in an internal validation set and 0.851 in an external cohort. The model detected early-stage COPD (GOLD I–II) with AUCs ≥ 0.940 internally and 0.861 externally. A separate 11-compound model graded GOLD stages with a micro-averaged AUC of 0.849. Shapley additive explanations provided patient-level attribution and chemically plausible features, supporting biological interpretability. These findings support breathomics as a scalable complement to spirometry for COPD case-finding and severity grading, and motivate prospective trials of breath-guided early intervention.