Time‐to‐transformation transcriptomic modeling identifies oral squamous cell carcinoma‐initiation and SPP1–integrin niche programs in oral leukoplakia

Oral leukoplakia and other oral potentially malignant disorders carry heterogeneous risks of subsequent oral squamous cell carcinoma (OSCC), whereas conventional clinicopathological factors have limited ability to discriminate individual risk and event timing. We developed a time-to-transformation transcriptomic score (TT-score) in the longitudinal GSE26549 cohort, including 86 participants and 35 subsequent oral-cancer events. A 12-gene TT-score was derived using multimethod feature selection and Cox-GLMNET modeling, and the full development pipeline was evaluated by 20 repeated five-fold nested cross-validations. Strict internal validation yielded a mean out-of-fold C-index of 0.740 (95% confidence interval, 0.679-0.801), with mean time-dependent area under the curve (AUCs) of 0.731, 0.771, and 0.811 at 2-, 3-, and 5-yr, respectively. Full-pipeline outcome permutation analysis supported the non-random nature of the observed risk-ranking signal. Across external progression-axis cohorts, direct stage associations of the fixed TT-score were cohort-dependent, whereas TT-score-related epithelial proliferation, invasion, hypoxia/metabolic, OSCC-initiation-like, and SPP1-related microenvironmental programs showed more consistent stage-associated trends. Single-cell analyses and published spatial-atlas signatures further contextualized these programs. Overall, the TT-score captured a transcriptomic risk state associated with the timing of subsequent oral cancer and provides a candidate framework for independent longitudinal validation and targeted assay development.