Bladder cancer (BC) remains a major public health concern, for which early detection is essential for improving patient prognosis. Emerging metabolomics research has highlighted the potential of urinary volatile organic compounds (VOCs) as non-invasive biomarkers for the early detection of BC. Cancer-related metabolic alterations give rise to distinct VOC profiles, facilitating the differentiation between malignant and non-malignant conditions. Advanced analytical techniques, such as gas chromatography-mass spectrometry (GC-MS), in combination with machine learning (ML) algorithms, offer a powerful approach for biomarker discovery and validation. This study aimed to leverage ML algorithms applied to GC-MS-based metabolomics data to discover a panel of volatile biomarkers able to accurately discriminate BC patients from cancer-free controls.

Urine samples were collected from 177 participants, including 87 patients diagnosed with BC and 90 cancer-free controls. This study was approved by both the Data Protection Officer and the Ethics Committee of the Portuguese Institute of Oncology of Porto (CES 82/022) and the Ethics Committee of the Faculty of Pharmacy of the University of Porto (CEFFUP). Urine sample analysis was conducted using headspace solid-phase microextraction coupled with GC-MS. Five distinct ML algorithms were employed for data analysis: support vector machine (SVM), random forest (RF), partial least squares-discriminant analysis (PLS-DA), extreme gradient boosting (XGBoost), and k-nearest neighbors (k-NN). Model performance was evaluated using receiver operating characteristics (ROC) analysis, with key metrics including area under the curve (AUC), sensitivity, specificity, and accuracy.

Among the various ML algorithms tested, the RF model demonstrated the highest performance in distinguishing BC patients from cancer-free controls. A panel of eight potential biomarkers was identified through RF for the overall detection of BC, achieving 89% sensitivity, 92% specificity, 91% accuracy, with an AUC of 0.872. This biomarker panel consisted of three ketones, three aldehydes, one fatty alcohol, and one phenolic compound. Of these eight compounds, seven were found at higher levels in the urine of BC patients compared to controls, while one compound was found at lower levels.

Our study identified a panel of eight volatile biomarkers that accurately detect BC. These findings underscore the potential of metabolomics, combined with ML techniques in cancer biomarker discovery. Implementation of this urinary biomarker panel in clinical practice may revolutionize BC detection by offering a non-invasive and rapid screening method.