Comprehensive Summary
This study by Ardila et al. aimed to synthesize diagnostic and prognostic performance metrics of machine learning based biomarker models in oral squamous cell carcinoma and integrated biological insights through a functional metasynthesis. Using a systematic review and meta-analysis, the authors combined molecular biomarker data with ML methods in oral cancer and obtained metrics such as accuracy, sensitivity, and specificity. Specifically, salivary DNA methylation biomarkers had an AUC of 1.00 in some models, which suggests perfect discrimination in those particular datasets. Overall, the review found growing evidence that ML models built on molecular biomarker data show promising diagnostic and prognostic capabilities in oral cancer. However, the authors mention that for many studies, external validation is lacking, and often the models are trained/tested on relatively small or single-centre cohorts.
Outcomes and Implications
This study concluded that applying machine learning to molecular and imaging biomarkers provides significant diagnostic and prognostic potential in OSCC. However, due to factors such as small sample sizes, limited calibration reporting, and limited external validation, the findings should be interpreted with caution. These limitations highlight both the promise of ML-based biomarkers for non-invasive early detection, risk stratification, and precision staging in OSCC, and the need for larger, multicentre, and standardized studies to enable clinical translation.