Abstract
Malignant pleural mesothelioma (MPM) is a rare and aggressive malignancy associated with occupational or environmental exposure to asbestos. Effective management of MPM remains challenging due to its prolonged latency period and the typically late onset of clinical symptoms. Accordingly, there is an increasing demand for the implementation of reliable, non-invasive, and data-driven diagnostic strategies within large-scale screening programs. In this context, the chemical profiling of volatile organic compounds (VOCs) in exhaled breath has recently gained recognition as a promising and non-invasive approach for the early detection of cancer, including MPM. Therefore, in this cross-sectional observational study, an overall number of 125 individuals, including 64 MPM patients and 61 healthy controls (HC), were enrolled. End-tidal breath fraction (EXP) was collected directly onto two-bed adsorbent cartridges by an automated sampling system and analyzed by thermal desorption-gas chromatography-mass spectrometry (TD-GC/MS). A machine learning approach based on a random forest (RF) algorithm and trained using a 10-fold cross-validation framework was applied to experimental data, yielding remarkable results (AUC = 86%). Fifteen VOCs reflecting key metabolic alterations characteristic of MPM pathophysiology were found to be able to discriminate between MPM and HC. Moreover, twenty breath samples from asymptomatic former asbestos-exposed (AEx) and eight MPM patients during follow-up (FUMPM) were exploratively analyzed, processed, and tested as blinded samples by the validated statistical method. Good agreement was found between model output and clinical information obtained by CT. These findings underscore the potential of breath VOC analysis as a non-invasive diagnostic approach for MPM and support its feasibility for longitudinal patient and at-risk subjects monitoring.