Development and Multi-center Validation of a Breathomics-Based Triage Tool for Lung Cancer: A Prospective Study of 5,214 Participants.

BACKGROUND: The inherent false-positive rate of computed tomography (CT) necessitates efficient, non-invasive triage tools to distinguish lung cancer (LC) from benign mimics, thereby streamlining early detection and mitigating unnecessary invasive procedures. RESEARCH QUESTION: Can a machine learning (ML)-derived-, molecularly-resolved breathomics prediction model effectively triage patients with radiologically detected pulmonary abnormalities in a real-world symptomatic cohort? STUDY DESIGN AND METHODS: In this large-scale, prospective, multicenter diagnostic accuracy study, we enrolled 5,214 symptomatic patients with radiological lung abnormalities at two campuses. Participants were allocated into a discovery cohort (n = 4,669) and a geographically independent external validation cohort (n = 545). Exhaled volatile organic compounds (VOCs) were analyzed using high-throughput proton-transfer-reaction time-of-flight mass spectrometry (PTR-TOF-MS). An ML model integrating a specific VOCs signature with clinical factors was developed, locked, and validated blind in the external cohort. RESULTS: The integrated model achieved an area under the receiver operating characteristic curve (AUC) of 0.891 (95% confidence interval (CI): 0.864-0.915) in the independent internal testing set. In the independent external validation, the model maintained robust performance (AUC 0.850, 95% CI: 0.805-0.890). In this validation cohort, applying the pre-specified 'rule-out' threshold, the model achieved a sensitivity of 93.1% (95% CI: 90.5%-95.4%), with an overall specificity of 55.9% and negative predictive value (NPV) of 71.0%. Importantly, in the intended-use pulmonary medicine subgroup (n=155), the sensitivity was 91.4%, with specificity and NPV reaching 55.8% and 89.0%, respectively. Subgroup analyses confirmed consistent efficacy across diverse clinical scenarios, notably maintaining robust accuracy in detecting early-stage disease (AUC 0.849). INTERPRETATION: This study represents the largest prospective validation of a mass-spectrometry-based breath test to date. The validated prediction model holds potential to serve as a robust, non-invasive triage tool. Its integration into the diagnostic pathway shows promise for enhancing efficiency and reducing unnecessary biopsies, particularly in respiratory outpatient settings.