Abstract
OBJECTIVES: Diagnosing post-traumatic stress disorder (PTSD) remains a challenge due to symptom variability and comorbidities. Linguistic analysis offers an innovative approach to identify PTSD symptoms and severity. This systematic review aimed at identifying linguistic features associated with PTSD, assessing the quality and limitations of existing studies, summarizing the predictive performance of identified models, and describing the clinical utility of these models. MATERIALS: A comprehensive search was conducted across multiple databases, resulting in the identification of 593 articles. After screening and eligibility assessment, 58 studies were included. METHODS: Data extraction focused on study characteristics, methodology, and performance metrics. We assessed the risk of bias using the PROBAST and conducted both a narrative synthesis and a meta-analysis. RESULTS: Linguistic features such as pronoun use, emotional valence, cognitive processing words, narrative length, discourse disorganization, temporal orientation, specific lexical fields (death, anxiety, sensory-perception details), and disfluencies were commonly investigated. The meta-analysis revealed a pooled area under the curve of 0.81, indicating the high performance of classification models. However, significant publication bias and heterogeneity were noted. Only 8 studies were rated with a low risk of bias, highlighting common issues such as inadequate control groups, unvalidated linguistic tools, unvalidated diagnosis tools, and low rigor in statistical analysis. DISCUSSION AND CONCLUSIONS: Linguistic markers showed potential for enhancing PTSD diagnoses, but the contemporary research was limited by methodological inconsistencies and biases. Future research should focus on standardized tools, symptom-focused studies, and interdisciplinary collaboration to improve the robustness and clinical applicability of findings.