Integrative prognostic modeling and mediation analysis of recurrence risk in extremely early-stage oral squamous cell carcinoma

BACKGROUND: Despite complete surgical resection, recurrence remains a substantial challenge in patients with early-stage oral squamous cell carcinoma (OSSC). Conventional TNM staging does not fully capture the biological heterogeneity underlying relapse. This study aimed to develop an individualized recurrence-prediction model integrating clinicopathologic, immune, and spatial factors, and to evaluate the potential mediating role of perineural invasion (PNI). METHODS: We retrospectively analyzed 451 patients with stage I-II OSCC treated at Xiangya Hospital between 2018 and 2025. Twenty-three clinical, pathological, and immunohistochemical variables were initially screened using LASSO regression. Independent predictors of recurrence were subsequently identified through multivariable Cox analysis. A nomogram was constructed based on the final Cox model and internally validated using Harrell's C-index, time-dependent AUCs, and 1,000-bootstrap calibration. Risk stratification was performed using Kaplan-Meier curves, and mediation analysis was used to determine whether PNI mediated the effects of tumor grade, P53 status, and age on recurrence. RESULTS: Six independent predictors were incorporated into the final model: tumor grade, PNI, postoperative lymphocyte nadir, tumor-to-midline distance, pathological stage, and P53 status. The model demonstrated good discrimination (C-index = 0.79) and strong time-dependent predictive accuracy (AUCs: 0.819, 0.825, and 0.807 at 1, 3, and 5 years, respectively). Calibration curves showed excellent agreement between predicted and observed recurrence probabilities. Risk stratification based on the nomogram clearly separated patients into low- and high-risk groups (log-rank p < 0.001). Mediation analysis showed that PNI partially mediated the effects of tumor grade (proportion mediated = 29.6%) and P53 status (7.7%) on recurrence, whereas age exhibited no significant mediation through PNI. CONCLUSIONS: This integrated prognostic model combines immune recovery, spatial invasion, and molecular features to accurately predict recurrence in early-stage OSCC. The partial mediation of tumor aggressiveness through PNI highlights a biological pathway linking tumor phenotype to recurrence risk. The proposed nomogram provides a clinically applicable tool for postoperative risk stratification and may assist in tailoring surveillance strategies and individualized adjuvant treatment decisions.