A Language Performance Model for Predicting Glioma Recurrence and Molecular Biomarkers: A Retrospective Cohort Study

Glioma progression is often accompanied by language dysfunction, and postoperative recurrence is nearly inevitable, especially in high-grade cases. This study aimed to identify valuable language-based prognostic markers and improve risk management strategies.

A retrospective longitudinal cohort of 191 glioma patients (2010-2018) was analyzed. Language status was assessed using the Aphasia Battery of Chinese (ABC), adapted from the Western Aphasia Battery. Principal component analysis (PCA) addressed collinearity in language scores, and Cox regression identified predictors for a survival model. Bootstrap validation and SHapley Additive exPlanations (SHAP) were used to confirm model stability and interpretability. Logistic regression was used to test associations between predictors and pathological molecular parameters.

Auditory verbal comprehension & writing and repetition emerged as key language predictors in the Cox model, achieving an AUC of 0.834. SHAP analysis confirmed their dominant contribution, defining the model as the Language Tests Combinations (LTC) model. Two-sample t-tests revealed significant differences in language scores between IDH1/2-mutant and wild-type groups. Logistic regression showed strong correlations between language predictors and molecular parameters (MGMT codeletion: AUC = 0.922; 1p/19q: AUC = 0.813; IDH1/2: AUC = 0.748), with tumor locations included as dummy variables.

Language components were identified as robust predictors of glioma recurrence and were significantly associated with key molecular features. The LTC model represents an internally validated and interpretable exploratory prognostic framework that may complement existing risk-stratification approaches and inform future studies on postoperative glioma management.