Abstract
Early detection of oral squamous cell carcinoma (OSCC) significantly enhances treatment outcomes and survival rates, with lymph node metastasis serving as a main prognostic factor. However, current clinical practices rely on TNM classification, including histological confirmation of metastatic disease in lymph nodes, often involving elective neck dissection, a procedure that can cause post-operative morbidity. Here it is shown that zinc imidazole framework-8 (ZIF-8) electrochemical biosensors can effectively distinguish non-metastatic (N0) from lymph node metastatic (N+) OSCC saliva samples. By monitoring the OSCC biomarkers cystatin B (CSTB), leukotriene A 4 hydrolase (LTA4H), and collagen type VI alpha 1 chain (COL6A1) in human saliva through electrochemical impedance spectroscopy and antigen-antibody immunoreactions, elevated biomarker levels in N0 samples are observed. The biosensor displays high accuracy, specificity, and reproducibility, with limits of detection lower than 0.4 ng mL(-1). Supervised bioinformatic analysis, using 34 machine learning classifiers, indicates LTA4H as the most accurate biomarker for distinguishing prognostic groups, confirming previous mass spectrometry findings. Notably, the AdaBoost model, integrating the combined detection of biomarkers, achieves a 76% accuracy rate in identifying metastatic saliva samples. This non-invasive biosensor technology, combined with bioinformatics, presents a sensitive and reliable approach to improve clinical assessments and guiding therapeutic decisions for OSCC patients.