Answers are grounded in this paper and cite their sources.
Background/Objectives: Formal thought disorder (FTD) is a clinically important feature of psychosis, but its assessment is time-intensive and dependent on trained raters. We hypothesized that a Thought and Language Disorder (TALD)-guided large language model (LLM) framework could identify clinically meaningful FTD features and distinguish healthy controls, clinical high-risk (CHR) non-converters, and CHR converters. Methods: The Story Game transcripts from 74 adolescents and young adults were analyzed, including 15 healthy controls (HC), 40 CHR non-converters (CHR−), and 19 CHR converters (CHR+). A modified 16-domain TALD framework was operationalized using a two-stage LLM agent that screened utterances and assigned domain-specific severity scores. Group differences were examined using sex-adjusted regression models with false discovery rate correction. Ten classifiers and seven feature sets were evaluated using repeated stratified five-fold cross-validation and leave-one-out cross-validation. Results: Significant sex-adjusted group effects were observed for Dissociation of Thinking, Poverty of Thought, and Verbigeration, which also ranked highest in the consensus feature-importance analysis. Classification performance was strongest for HC versus CHR (balanced accuracy = 0.881), followed by three-class classification of HC, CHR−, and CHR+ (0.753), and CHR− versus CHR+ classification (0.743). The three-class model exceeded chance performance in permutation testing (p = 0.005). Conclusions: A TALD-guided LLM framework identified a focused set of clinically interpretable FTD features associated with psychosis-risk status and transition. These findings support the feasibility of scale-guided, transcript-based FTD assessment while underscoring the need for validation in larger and more diverse samples.
No takes yet. Share an insight, caveat, or question.
Zhang et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: