NCOG-42. SPEECH-SPECIFIC ERRORS IN LEXICAL RETRIEVAL PREDICT GLIOMA MOLECULAR FEATURES

Abstract Recent work confirms that low and high-grade gliomas, remodel neural circuits which influence speech production. The extent to which speech kinetics and category of speech error during lexical retrieval may predict glioma molecular features remains unknown. Here, we extracted audiovisual speech production from 550 patients totaling over 26,400 individual stimuli to identify correlations with glioma molecular features including IDH, TERT, 1p19q co-deletion, and CDK2AB. Nonparametric statistics were used for analysis. IDH wild-type tumors were more likely to be temporoparietal, while IDH mutant tumors were more likely to be insular (p=0.02). Patients with IDH wild-type glioblastoma exhibited more errors (8.23 vs 3.79, p=0.02), with a higher occurrence of anomia (p=0.004), circumlocution (p=0.034) and general delay (p=0.04) when compared with IDH mutant WHO grade 4 astrocytoma. Binomial logistic regression identified lexical retrieval anomia, circumlocution, and initiation delays as an independent predictor of having an IDH wild-type tumor. Within the IDH mutant group, anomia (p=0.03) and articulation (p=0.014) errors were associated with malignant molecular features (after controlling for prognostic variables such as tumor volume). Binomial logistic regression identified articulation or anomia for picture naming as an independent risk predictor for WHO grade 2 gliomas with greater risk molecular features (based on Ki67 and CDK2AB). Distinct patterns of language errors were observed in patients with IDH wild-type and IDH mutant tumors. Further work is required to elucidate the functional and biological correlates of these findings.