Current research status of risk assessment models for oral squamous cell carcinoma based on microbiome databases

Oral squamous cell carcinoma (OSCC) is one of the most common oral malignancies. Due to insidious early symptoms, high invasiveness, and frequent cervical lymph node metastasis, patient survival remains poor. Recent studies demonstrate significant associations between oral dysbiosis and OSCC. To address limited early screening and suboptimal survival rates, this review systematically summarizes OSCC microbiome research and disease risk assessment applications. Studies reveal marked oral dysbiosis in patients with OSCC, characterized by enrichment of Fusobacterium and Porphyromonas , reduced Neisseria abundance, and altered community stability. These promote tumorigenesis through chronic pro-inflammatory signaling, the production of carcinogenic metabolites, and direct oncogenic pathway interference. Consequently, the oral microbiome offers non-invasive, sensitive biomarkers for early detection. We identify and distinguish four categories of predictive models in the current literature: (1) detection/discrimination of prevalent OSCC, (2) prediction of malignant transformation from oral potentially malignant disorders, (3) prediction of recurrence or metastasis after treatment, and (4) prognostic models for survival prediction. Existing databases include comprehensive and OSCC-specific collections with multi-omics integration. Risk assessment models encompass traditional statistics, machine learning, deep learning, and multimodal fusion, and have shown promising performance in retrospective cohorts. However, standardized databases are lacking, and existing models face confounding interference, limited generalizability, and insufficient interpretability. Furthermore, current predictive models are predominantly based on 16S rRNA sequencing (bacteriome-centric), whereas comprehensive risk assessment requires integrating interkingdom entities, including fungi, viruses, and archaea, through metagenomic and virome profiling. Critically, the distinction between promising retrospective discrimination and validated prospective clinical utility remains a major translational barrier. Future directions include developing standardized longitudinal databases, integrating multi-omics, using explainable artificial intelligence, and conducting prospective validation studies to achieve personalized precision treatment.